Closed-loop AI Key to Transformational Impact

Leaders of process manufacturing industries confirm in various studies and surveys that the potential impact of artificial intelligence (AI) on their business is indeed transformational.

Double-digit production efficiency gains and maintenance cost reduction are often cited. Some state that they are already seeing some of the benefits and returns. 

For those who have seen the benefits, supply chain and logistics optimization and predictive maintenance are typically mentioned as the key sources of gains from investment in AI. 

However, for process manufacturing the most significant area of impact, appearing on the top and the bottom line, is the reduction in the cost of conversion and an increase in production efficiency.

In increasingly global and open markets where most other market factors are normalized, control of conversion cost and production efficiency become the only two levers for differentiation and competitive edge.

Yet, the present use of AI for advanced data analytics, and, at best, some form of prediction, prognostic, or recommendation does not impact conversion cost or production efficiency in a significant way. Without this impact and attendant ROI, AI is therefore not transformational. 

Of course, the ultimate prize is transformation impact, which can be achieved by letting AI take control, ultimately closing the loop between analytics and action. This outcome is called closed-loop AI.

Relying on humans (open-loop AI) to take the final action (even with well-implemented, AI-based predictions) eventually leads to AI not getting used at all, is an obstacle to widespread use, and cannot lead to sustainable ROI.

Open-loop injects delay, uncertainty, and subjective decision-making into the process, undermining the goals to be objective and data-driven. 

Closed-loop AI Continuously Improves Outcomes

AI algorithms deployed initially will not be perfect but running them in the open loop will not provide any opportunity for improvement. This can be compared to running all your PID control loops manually because they are poorly tuned.

By using closed-loop AI, the initial deployment costs of machine learning can be reduced. Additionally, the insight generated by the algorithms is continuously optimized as the machine learning improves with data and experience. 

Achieving real benefits in productivity gains and building flexible manufacturing supply chains needed for the future requires closed-loop AI.

Today, the needs of the supply chain are translated by humans or middleware and some form of Process Control System (PCS) is used to orchestrate control strategy at the plant floor.

In contrast, a Value Control System (VCS) can be built with AI and digital twins to make the plant floor equipment respond directly to the dynamic needs of a flexible supply chain.

APC is a Logical Starting Point for Closed-loop AI

The proposition of closed-loop AI often evokes fear and uncertainty. Security, latency, and lack of trust in AI are typically the most common barriers cited.   

At the heart of these apprehensions is the misconception of what a “loop” is in this context. A loop is not a sensor-to-actuator loop with a PID or logic algorithm, but a number of value control loops optimizing constraints in a process cell, process unit, or entire process plant. 

The use of constraint-optimization control techniques such as MPC has gained significant momentum in the last two decades. Some progressive and advanced users apply MPC widely in lieu of PID control.

For users already familiar with these advanced process control (APC) techniques, closed-loop AI for APC is a logical step. They can and should see immediate incremental benefits from using AI.

Initial implementation can be started by using AI to supplement traditional APC to overcome some of the challenges that make MPC ineffective or hard to maintain. This hybrid approach of combining process dynamics and first principles-based models with machine learning models can be the most effective way to start building the VCS for the future.

The better tools the platform brings help overcome challenges of current APC techniques.

Rajiv Anand – Quartic.ai

Modern software platforms now allow industrial users to build AI applications, without the need to have data science skills to do so. The better tools the platform brings help to overcome the challenges of current APC techniques.

Additionally, role-based user interfaces also give process control professionals the experience and confidence to use AI in process optimization and control.    

AI applications built by these subject matter experts, just like any other APC will make adoption of AI and the related benefits widespread and mainstream.

Rajiv Anand – CEO Quartic.ai

AI systems are becoming available that automate many of the machine learning tasks that could previously only be performed by experienced data scientists. Using these OT-focused AI platforms, engineers and manufacturing subject matter experts can widely apply AI to develop these next-generation value control systems.

AI applications built by these subject matter experts, just like any other APC will make adoption of AI and the related benefits widespread and mainstream. Highly interpretable and explainable AI is also becoming available to increase trust and confidence in closed-loop AI.

Behind the Byline

This blog was originally published on https://www.arcweb.com/

Rajiv Anand is the co-founder and CEO of Quartic.ai. The company provides a smart Industry platform for process manufacturing industries that are implementing digital transformation with AI and IIoT. 

Rajiv has more than 30 years of experience in process control and asset health solutions using Emerson platforms for various process manufacturing industries. Rajiv held key engineering, and leadership positions with Emerson and their Impact Partners. 

Stay in touch