Improve the robustness of your specification drift control with AI

Improve the robustness of your specification drift control with AI

Process manufacturing is susceptible to what is usually referred to as specification drift or spec drift. Whether the process uses extruders, chemical vats, or any other continuous process, control and command systems can’t always ensure the final manufactured product doesn’t drift from its expected specifications.

Why does drift happen? The reasons are multiple and not always under an operator’s control. For instance, variability in the plastic pellet, cornmeal water-absorption properties, plant temperature or humidity, or even equipment behavior based on wear or tear can impact the overall manufacturing process.??

Traditional control systems and spec drift?

A traditional control system, such as a PID controller, to name a popular one, will often control manufacturing parameters based on an expected target. It will not control the process based on an expected outcome.??

For instance, the PID will control the rotation speed of an extruder to ensure it stays at the desired rate. This rate should, in turn, guarantee that the extruded product has the appropriate characteristics. The keyword here is “should.” If, for instance, the plastic rod is too long coming out of the extruder, an additional mechanism will need to be put in place to modify this rotational speed – and any other parameters – to ensure that the rod stays within its expected specifications. These mechanisms are often linked to an operator’s expertise and ability to adapt control systems setpoints to ensure output quality.?

What if, instead of solely relying on operator expertise, we could develop an AI agent that could harvest their knowledge and help all operators control specification drift, regardless of their experience level? What if a “meta-controller,” an AI-driven one, could control the traditional control systems??

Improving robustness with AI agents?

The example below depicts the variability of specific food product specifications coming out of an extruder. As it illustrates, despite the precisely controlled extruder and oven, several product characteristics evolve and would go beyond their tolerance band if nothing was done.?

No alt text provided for this image

How did the food manufacturer ensure that the production consistently stayed within its approved tolerance band????

To achieve this result, they used a relatively novel advanced AI approach. With this approach, an AI agent is trained using deep reinforcement learning, aka an?autonomous system. This agent will serve as a “meta controller” to control the control systems (directly through, e.g., the PLC or through a SCADA) and adapt target set points. It will ensure that the overall output stays within its given specifications.?

No alt text provided for this image

With this approach, the AI agent – aka the “brain” in the?Microsoft Project Bonsai toolchain?vernacular- can overlook the total system holistically.??

  • It?monitors?both the different manufacturing equipment as well as the output specification itself (for instance, a plastic rod length or the color or shape of a food snack)?
  • It allows the agent to?dynamically change?control parameters to ensure the final output stays within its specifications?
  • It?catches drift?before the product goes out of its tolerance level, and potentially large amounts of product will need to be thrown away or, in a best-case scenario, has to be reused or recycled??

Besides managing specification drift, DRL-trained AI agents can help process manufacturing in various ways. They can leverage deep neural networks and simulations to help improve processes in a way not attainable with traditional command and control systems.?

Next steps?

Please refer to these articles to learn more about how autonomous system AI agents trained using Microsoft Project Bonsai can help control specification drift.

?

(This article was originally published on Neal Analytics blog)

要查看或添加评论,请登录

社区洞察

其他会员也浏览了