The path from PGD to autonomy


When you look at the history of automation, it’s not hard to see that most of the advances are extensions of existing technologies. This can be seen at all levels of automation – from the evolution of relays to programmable logic controllers to the company where, in the course of the course of studies, holistic ERP (Enterprise Resource Planning) systems were developed from extensions to material requirements planning (MRP) over several years.

However, when looking at autonomous systems, the progress can seem so great that it is not always so easy to identify the origin in traditional manufacturing systems. This loophole can lead end users to become suspicious of autonomous technologies; But if you follow the evolutionary path of autonomous systems, these new technologies can be less intimidating.

Connect the dots
at Rockwell Automation At the 2021 trade show, Jordan Reynolds, Global Director of Data Science at Kalypso (a Rockwell Automation company), gave a presentation on autonomous systems that helped explain how these advanced neural systems, as used in industry, are extensions of the Closed-loop PID (proportional-integral-derivative) control systems that we are all familiar with.

Jordan Reynolds, Global Director of Data Science at Kalypso (a Rockwell Automation company).To illustrate the evolution from PID control to autonomous systems, Reynolds explained that the first step is to start with a physical system – an entire plant or a production line – and create a model or digital twin of that system that shows how this system reacts to changes in inputs or operating parameters as well as disturbances.

This model, which creates the conditions for autonomy, is created through hybrid modeling. Reynolds said that hybrid modeling is developed through two processes: first, there is input from an engineer, followed by input from a data scientist who understands AI (artificial intelligence).

“There’s a delta between what an engineer knows and what a data scientist does to derive a model that can learn instead of being programmed,” Reynolds said. He stressed that in order to develop an effective model for autonomous operations, the engineer must define the basic principles and let the data scientist fill in the gaps to ensure that the model complies with the standards. “This is hybrid modeling,” he said.

Autonomous control
Rockwell Automation focuses on this area because it views autonomy problems as control problems.

“Feedforward control was one of the first predictive uses of the control,” said Reynolds. “It expands the feedback control and provides early information about a pending condition so that it can be proactively addressed. Model Predictive Control (MPC) is a modern version of feedback control in which you use multivariable models to characterize the performance of a system. We use these models to control a system better than we could with PID. “

Reynolds noted that Rockwell Automation acquired Pavilion Technologies for its MPC technology in 2007.

“MPC works well with highly predictive systems that don’t change as much,” Reynolds said. “But it falls short when recipes or parameters change. The MPC needs to be updated as it does not adapt by itself. This is where adaptive control comes in. With adaptive control, the model doesn’t have to be perfect because it can adapt to changes. “

Reynolds' presentation included a slide that explained reinforcement learning in an industrial production construct.Reynolds’ presentation included a slide that explained reinforcement learning in an industrial production construct.Reinforcement learning is a newer concept that extends adaptive control and is an emerging method of AI that has evolved along with the capabilities of cloud computing resources and greater connectivity between manufacturing systems.

Explaining the connections between PID and autonomous operations using reinforcement learning, Reynolds said that PID is used to ensure the system is meeting setpoints, while MPC determines setpoints based on the system’s multiple inputs and outputs. In addition, reinforcement learning creates an executive function that can develop strategies.

Reynolds provided an example of auto racing to illustrate this concept. MPC keeps the car on track, he said, while reinforcement learning helps you develop strategies on how to win the race, such as:

Deep learning neural networks, such as those used in reinforcement learning, are where most of the innovations have emerged as the industry advances towards greater use of autonomous systems. The problem is that neural networks are very precise, but not very transparent or explainable. “An engineer can’t just get an answer to why decisions are made by the neural network,” Reynolds said.

For this reason, Rockwell Automation focuses on deep symbolic regressions to explain how an AI control model works. Reynolds said Rockwell Automation did this in its FactoryTalk Analytics LogixAI. FactoryTalk Analytics uses LogixAI analytics within the control application to drive process improvements, according to Rockwell Automation. It is an add-on module for ControlLogix controllers that streams control data over the backplane to create predictive models.

Source link


Leave A Reply