Use predictive analytics in manufacturing to gain insight

Use of predictive analytics in manufacturing

Predictive analytics, which are an essential part of a predictive manufacturing system, can provide more insight than ever before into the causes of manufacturing issues. This is more information about problems than predictive maintenance alone is able to give. In fact, a simple way of distinguishing the two is that predictive maintenance focuses more on plant and equipment, while predictive manufacturing is more interested in the end product.

And manufacturing is currently leading the way when it comes to utilizing predictive analytics, said Mike Leone, an analyst at IT strategy firm Enterprise Strategy Group. For example, manufacturers can gather and use IoT data during the predictive analytics process in a connected factory, then use that to automate processes with real-time monitoring and alerting. Companies can also use predictive analytics to track packages, manage inventory, ensure worker safety and optimize the supply chain.

The information gleaned from a predictive manufacturing system can also influence everything from supplier selection to design. It can affect decisions ranging from day-to-day judgment calls to what the manufacturer decides to produce.

And it's crucial that that software collects data during each step of manufacturing.

Predictive analytics in manufacturing relies on collecting sensor data across the manufacturing process, Leone said. If this happens, manufacturers can uncover trends, forecast outcomes, improve and ensure product quality and optimize asset allocation and capacity utilization.

In addition, a predictive manufacturing system doesn't just come in handy during one step of the manufacturing process.

A predictive manufacturing system has many roles in a factory, said Forrester analyst Paul Miller. For example, it can help when a manufacturer is striving for reliability or searching for the most cost-effective energy mix or the ideal materials. 

"Potentially, for a complex industrial asset, there could be thousands and thousands of combinations to consider," Miller said.

However, adopting a predictive manufacturing system doesn't come without its growing pains.

A key barrier to adopting predictive manufacturing is internal resistance, Miller said. 

"People will say, I have been running this operation for 30 years and I know how to do it," Miller said.

However, some companies have been pleasantly surprised by the results, he said. 

"Siemens [an automation company] found [that] with their gas turbines, they were able to get significantly better performance than their best engineers because the computer can try so many options all at once," Miller said.

Using a predictive manufacturing system has plenty of other applications as well.

For example, companies can use it when the supply chain lacks consistency, Miller said. For example, a company might be getting metal castings from different foundries or a power station has to operate with coal sourced from different mines.

"Those inputs have different properties, but predictive can help change the configuration that depend on them and still get efficiency or the right cost mix or quality," Miller said.

Growing interest in predictive analytics

For many manufacturers, the predictive analytics that a predictive manufacturing system offers could help solve many different problems.

"The drivers for applying predictive will be different depending who you are," Miller said. "Sometimes the focus is on cheapness or quality or using the least energy, and for each, there are levers you can tweak to get that."

While the pandemic has changed many things about the world, the requirements for manufacturers aren't one of them. A predictive manufacturing system can make it easier to meet those requirements.

Prior to COVID-19, there was pressure on manufacturers to be as reliable and predictable as possible, Gartner analyst Simon Jacobson said. Those imperatives are still there, and, if anything, have been garnering even more attention. However, predictive manufacturing interest isn't all pandemic-driven.

The flurry of recent predictive implementations is still anchored by different analytic techniques, whether that's looking at multi-variate statistics or taking part in pattern analysis or forecasting, Jacobson said. Two factors have helped to boost interest recently: a reduction in the cost of computing needed to work with large data sets and the rise of a self-service model that doesn't depend on direct work by a data scientist.

"You aren't necessarily looking for a needle in a haystack anymore -- you are just going after a better view of the haystack," Jacobson said. 

To determine that root cause [of an issue] is one thing, but to make sure it doesn't happen again is where predictive comes in.

For example, an organization might find there's an issue with a specific variability in a batch of product, and the organization needs to determine whether it's a raw materials issue or something else that caused that variability, he said.

"To determine that root cause is one thing, but to make sure it doesn't happen again is where predictive comes in," Jacobson said.

If a manufacturer has a predictive manufacturing system, it can look at all the data -- everything from materials to the habits of particular workers to an increase in humidity within the plant – and decide what factors to monitor going forward to prevent a repeat of the problem.

Drawbacks of predictive analytics

Using a predictive manufacturing system and the predictive analytics that underpin it has become increasingly popular. That doesn't mean that there aren't potential problems that a manufacturer should be aware of before deciding to adopt one.

For example, one factor driving an "infatuation" with predictive analytics is the rampant experimentation with machine learning that's occurring in manufacturing -- and infatuation can be dangerous, Jacobson said. 

When manufacturers are gathering large volumes of data, they need to be sure that the data being captured is actually relevant, he said. In other words, data that correlates but isn't actually causative could lead a company to an incorrect conclusion, which could be costly or even potentially dangerous.

Manufacturers shouldn't blindly place their faith in a predictive manufacturing system.

"A lot of organizations have high confidence in machine learning algorithms but don't really have all the data, and that is a risk," Jacobson said.

For example, a product could become tainted and an organization may not have the algorithms to detect that that occurred.

If a predictive manufacturing system leads to a problem, that could tarnish the company's image.

"Machine learning and predictive gone bad is bad for brand equity," Jacobson said. "The unfortunate thing is that incompleteness in models does not seem to impair decision-making confidence."

New technology infatuation has given much of the industry a tendency to focus on positive outcomes rather than assessing machine learning and predictive manufacturing systems more critically.

"People treat data as a sacred version of the truth," he said.

Use of predictive analytics in manufacturing a journey

The key is understanding the nuances of a predictive manufacturing system and how best to use it to gain advantages.

For example, organizations that are striving for deeper understanding can gain that by using a predictive maintenance system, Jacobson said.

"I know of one chemical company that combined predictive with optical inspection and they used it to identify products that could be downgraded rather than simply rejected," Jacobson said.

Because of this, he said, the company was able to recover one to two million dollars a year because it was creating saleable product rather than waste.

"It sounds like just a cliché, but it really is true: Predictive is a journey," he said.