Control theory can optimize manufacturing?

I had a provisional patent on this idea, but I'm not going to spend any of my own money trying to patent the idea fully, and I can't find anyone to buy the idea or pay for the patent fees-- so here's just some ideas that I hope are helpful. I think they could be really, really useful in industries where there is extensive monitoring and control of manufacturing processes.

It all started with some conversations with my data engineering husband who had exquisite data, which led to this paper. In said paper, I showed that you could use an AI (or other) model that predicted manufacturing output and quality from the state of the manufacturing control knobs, whatever those might be, to optimize the manufacturing process. Imagine you have a model that takes in manufacturing control knobs and produces an accurate prediction of the output, with whatever metrics you care about. Training the model means that you maximize fidelity between learned and actual relationships from manufacturing control knob settings to manufacturing output metric. But optimizing the manufacturing process means changing the control knobs so that you, with this learned model, maximize the output metrics.

For instance, in pharmaceutical manufacturing, you might try to optimize batch yield. Control knobs might include how much sugar you feed the cell, the temperature setting, all that jazz. And so first, you'd build a model of how batch yield varies with control knob settings. Then, you'd tune control knob settings so that the model predicts maximum batch yield. One caveat is that you have to stay in the regime where the model will be accurate, so you can't go too far outside your training examples.

You can do even better than this if you used a closed-loop control scheme. Imagine now that you continuously read out the state of the system and also can continuously exert some control over the system. For instance, suppose you can read out the density of cells continuously and also can continuously change the temperature. Then, you can use closed-loop control methods instead of open-loop control methods to optimize the manufacturing process. Closed-loop control methods are more powerful than open-loop control methods, since you're using feedback to adjust your prediction of what you should do. Open-loop control settings might provide valuable guidance, but with any standard closed-loop control technique, you might eke out extra gains.

Every single gain in manufacturing output could save tons of money. Imagine a 10% increase in batch yield in pharmaceutical manufacturing. I can only imagine what that would lead to in terms of cost savings.

Below is a schematic for how you could implement control theory in a pharmaceutical manufacturing setting to optimize the manufacturing process. It could be adapted to any other industry easily.

If implemented well, this could be a huge part of Industry 4.0, the manufacturing-specific version of the Fourth Industrial Revolution. I bet someone in China is already using this idea.

Have at it, if you wish.