Nostal6ie/iStock
Factory in plant

Feedback Loops Build a Smarter Plant

Nov. 21, 2022
Data integrity is critical to achieving the right outcomes.

In an increasingly “smart” world, data is being created everywhere. Engineers and manufacturers generate massive amounts of data for every part, process and product made. However, the future of manufacturing is not centered around creating data but rather harnessing that data to improve designs, workflows and outcomes.

One method of leveraging data is to integrate feedback loops within the manufacturing process. Feedback loops are the fundamental mechanisms that enable individuals and computers to learn and make increasingly intelligent decisions over time. Feedback loops can drive continuous improvement in all parts of the engineering and manufacturing process, regardless of whether the process utilizes expert manufacturers or machine learning models.

To facilitate feedback, structured data collection and analysis make it easier to identify potential design issues, create better predictive models for manufacturability and cost, build more efficient workflows and steadily refine manufacturing outcomes. Creating the right models and feedback loops is key to solving almost any manufacturing problem.

Start With Clean, Organized Data

The first step towards smart manufacturing is standardizing the data collection and organization process. Engineers looking to leverage manufacturing data should thoughtfully consider structured data organization that functions for multiple applications and implement feedback loops that can provide a continuous stream of data to their models, either for continual learning or validation. Data standardization and organization are often overlooked; however, they are the foundation for developing manufacturing intelligence.

While planning the data structure, consider how the data is being recorded, who is recording it, how to address missing data and how to measure data accuracy. Data integrity is critical, and it is far better to develop intentional frameworks and guidelines from the start than to try to clean up poorly recorded, disorganized data after collection. Spending time to preemptively plan and organize your data will prevent future frustrations, and ultimately save time in developing sound feedback loops and machine learning models.

Automate Machine Learning

With a clean, structured data set, engineers can begin to implement feedback loops and computational models to solve a broad range of manufacturing problems. A machine learning model is an example of manufacturing intelligence that requires highly structured and consistent data. At the outset, that structured data is critical for properly training the machine learning model, but over time, prediction accuracy can degrade as training data becomes outdated. Feedback loops can automatically provide models with a continuous stream of new data to improve performance and enable automated retraining to correct for prediction bias, drift or inaccuracies.

Spending time to preemptively plan and organize your data will prevent future frustrations, and ultimately save time in developing sound feedback loops and machine learning models.

What does this look like in practice? Consider a manufacturing cost model as an example. Engineers are familiar with many of the cost drivers for manufacturing their designs, including material costs, labor, the manufacturing method used and the geometry of their part. With carefully structured data, it is possible to use machine learning to predict the cost to manufacture a part.

For each part manufactured, the actual cost of the part can be returned to the model to complete the feedback loop. This new data can then be integrated into the current testing and training data sets such that, over time, the model can steadily increase prediction accuracy. Most importantly, all of this can be done with little or no human intervention.

First Step: Manual Feedback Loops

While not every organization is ready to implement fully automated learning software, the benefits of feedback loops can still be realized with manual processes. Similar to how computational models learn through feedback data, feedback loops provide critical information for engineers to learn and understand how to improve processes, designs and estimates.

With accurate, structured data, engineers and analysts can develop many of the same insights and improvements as automated learning. In addition, maintaining human oversight in the feedback process enables organizations to rely more heavily on the expertise and experience of their engineering teams.

For example, a manufacturing cost model based on either machine learning or a deterministic equation can be manually updated when engineers observe unacceptable inaccuracies. As new data points and feedback are generated, the model can be routinely refined and adjusted manually by an engineer who has been educated with the feedback.

Whether using machine learning or a simple equation, data collected using a feedback loop is critical to informing the manual updates and accuracy improvements. While the manual approach may require more time and individual effort, it provides an effective stepping stone to an automated learning process. It also keeps the control in the hands of the engineers, which can help build trust and serve as a bridge to automation.

A Future of Incremental Improvement

Feedback loops are a cornerstone for growth and improvement throughout the manufacturing industry. The Fourth Industrial Revolution has given engineers access to an unprecedented amount of data, yet that data remains largely underutilized. As manufacturing centers transition into smart manufacturing facilities and data become more integral in daily operations, feedback loops and structured data sets will be critical to success.

Smart manufacturers today are implementing Industrial Internet of Things (IIoT) technologies, collecting more data points than ever, and actively searching for new ways to make products stronger, faster, more sustainably, and cheaper than ever before. With accurate data and analytical models that get better every day, the opportunities are endless.

Davis McGregor, Ph.D., is a senior manufacturing scientist with Fast Radius.

About the Author

Davis McGregor | Senior Manufacturing Scientist, Fast Radius

Davis McGregor, Ph.D., is a senior manufacturing scientist with Fast Radius.

Sponsored Recommendations

The entire spectrum of drive technology

June 5, 2024
Read exciting stories about all aspects of maxon drive technology in our magazine.

MONITORING RELAYS — TYPES AND APPLICATIONS

May 15, 2024
Production equipment is expensive and needs to be protected against input abnormalities such as voltage, current, frequency, and phase to stay online and in operation for the ...

Solenoid Valve Mechanics: Understanding Force Balance Equations

May 13, 2024
When evaluating a solenoid valve for a particular application, it is important to ensure that the valve can both remain in state and transition between its de-energized and fully...

Solenoid Valve Basics: What They Are, What They Do, and How They Work

May 13, 2024
A solenoid valve is an electromechanical device used to control the flow of a liquid or gas. It is comprised of two features: a solenoid and a valve. The solenoid is an electric...

Voice your opinion!

To join the conversation, and become an exclusive member of Machine Design, create an account today!