(159a) How AI Brings Precision Manufacturing to Chemical Processes

In this case study presentation, learn how a food manufacturer has implemented AI to accurately predict when the production of its animal feed additive product was completed in order to reduce batch-to-batch variability and maximize asset utilization.

The customer is a Fortune 100 Food Production company, serving clients in more than 160 countries. With a global value chain that includes over 400 crop procurement locations, 250 ingredient manufacturing facilities and numerous innovation centers, the company’s products are used for food, animal feed, industrial and energy uses. As part of their strategy to improve profitability, one of the company’s Ag Processing business units was looking for ways to reduce costs and increase yield in order to improve profit margins for a particular animal feed additive. However, this is a process whereby the yield and production time is vastly impacted by a number of variables that fluctuate depending on the operating conditions, process knowledge and individual operator. The business unit wanted to leverage predictive analytics to reduce the variability in the process, and ultimately lowering production costs and increasing asset utilization.

The production process for this specific animal feed additive being produced can take hours. The high-volume/low margin product that is subject to a highly complex and dynamic production process due to the variability that biological processes and operating conditions can have on it. Previously, the customer was using traditional methodologies to understand the behavior of the production process and to make decisions. But it found that traditional methodologies were limited in their ability to provide an accurate and repeatable process for creating, generating, and retraining data models. For example, Model Predictive Control (MPC) requires a model to be developed by process experts with a deep understanding of the relevant process changes, variables, and interactions. MPC is also very hands-on, whereby a subject matter expert needs to manually analyze a single set of linear models to make predictions over a fixed time horizon. When the data drifts or conditions change, the model then needs to be revised or rebuilt to accommodate new variations. This labor-intensive approach coupled with a lack of accurate information meant that the decision to stop the production process was often based on the experience of the individual operator, often leading to the process running for longer than necessary.

The customer was looking for an automated solution for modelling the process in order to accurately predict when production was completed, reduce batch-to-batch variability and maximize asset utilization.

The manufacturer’s process engineers used the Canvass AI platform to create a predictive model that would predict process behavior using data from the different OT systems as well variable lab data. Using the high frequency data input, the Canvass platform was able to generate data models that predicted production output over pre-determined time intervals. As a result, the client can visualize the actual and predictive production levels together, arming its operators with the necessary insight to quickly understand the status of the process and make decisions in real time.

By optimizing this specific production process, the customer estimates that overall asset utilization has increased significantly. The increased asset utilization means that the customer can now produce more of this specific animal feed additive without additional capital expenditure, or vice versa, and produce the same amount of the product at a lower cost. Further, the customer sees the opportunity to save additional costs by leveraging the findings from the AI-enabled predictive models to better manage predictive maintenance programs, yield improvements and energy utilization.

Going forward, the client will no longer need to manually analyze its process data. By taking the data from the OT systems and automating the entire data analytics process, the Canvass platform creates models powered by artificial intelligence that continually adapt to data changes and update insights in real-time. This ensures that the customer’s operational teams have the most up-to-date insights to accelerate their decision making.

This presentation will show that AI-powered predictions are being utilized to manage chemical processes today. The audience will learn that how to overcome the key barriers to AI success and why the three P’s: people, process and AI product are critical to success.