Batch processes and continuous processes are different.
So why do many machine learning solutions treat them the same?
Designed to work with industrial data, Fero's explainable machine learning-powered software has distinct ways to process batch and continuous data as it exists in your historian. In this post, you'll see how Fero tackles the unique challenge presented by each process type and drives millions of dollars in savings as a result.
How Fero optimizes batch processes
For manufacturers that produce goods in batches, keeping track of individual batches as they progress through production stages can be a daunting challenge. With huge volumes of data accumulating in your historian, it's difficult just to keep tabs on what's happening on the factory floor, let alone derive insights from that data.
It's reasonably easy to analyze single-stage batch processes. With Fero, you can analyze multi-stage batch processes—providing crucial beginning-to-end insights about your production—with simple definitions and no code.
You'll also get predictions and recommendations individualized for each batch in real time, driving huge increases in profits and efficiency.
One large chemical plant recently used Fero to optimize a thermal dispersants unit, with two goals in mind: understanding quality variation in their process, and demonstrating the business use case for a machine learning-driven optimization workflow.
The result? The plant improved a key quality metric and estimated up to $6 million in increased revenue as a result.
The team began by applying Fero in static mode—similar to an offline Six Sigma-type analysis, but using cutting-edge explainable ML methods that allow for root cause analysis and optimization. In static mode, the software achieved high predictive accuracy for several quality KPIs at the maleation, stripping, and amination stages of production, focusing on turbidity and viscosity as key quality metrics. As a bonus, the team was also pleased that Fero seamlessly integrated into their existing IT stack through an API, allowing for ease of use by operators.
Once confident in Fero's predictions, the team could switch over to live optimization, allowing them to analyze real-time data. Starting at the maleation stage, Fero could predict finished batch properties in real time, and recommend adjustments based on the current production parameters in order to optimize the quality of the finished dispersant. Complete batches could also be analyzed by Fero, further improving product quality in future.
How Fero optimizes continuous processes
In continuous production, the data looks completely different, and so do the challenges. Rather than tracking individual batches, you need to monitor the ongoing flow of product through the plant. That means you need a solution that can track how fast your product is moving, and understand how long it spends in different stages.
Fero users can customize the machine learning models by easily entering this information into the software—telling it, for example, how long a liquid molecule remains in each stage. This ensures the models are as accurate as possible, so the process can be optimized to minimize waste and maximize yield.
Facing a complex 18-stage continuous process with varying residence times, one major food manufacturer recently used Fero to solve two key challenges with that process: understanding and eliminating product quality inconsistencies between regions, and reducing end of line tests that sucked up the team's time without adding value. In static mode, Fero predicted 100% of the out-of-specification events that had happened earlier in the year; when live-optimized, the models reduced end-of-line tests by 40%. The team also saw increased throughput and reductions in energy and water waste, helping them satisfy multiple crucial KPIs.