Nst Industry Week 1540x800 Oct 2020

Don’t Believe the Myths About AI in Manufacturing

Oct. 1, 2020

The age of artificial intelligence (AI) in manufacturing has arrived, and it’s more accessible and affordable than many small- and medium-sized manufacturers (SMMs) realize. But like many technologically disruptive innovations before it, adoption of AI is slowed by a lack of awareness and concerns about costs, scalability and security.

The basis of AI is machine learning. Machine learning is the idea that through the continuous ingestion of data, and through iteration and refinement, machines actually “learn” or get “smarter.” AI refers to the use of this ever improving information where machines can perform tasks with “intelligence.” Think of machine learning as the brains of the operation, with AI providing a recommended action. The more that is learned about the data, the better the recommended actions get.

AI is most commonly applied in manufacturing to improve overall equipment efficiency (OEE) and first-pass yield in production. With experience, SMMs can use AI to increase uptime and ensure consistent quality, which makes for better forecasting.

AI implementation can seem a bit overwhelming. Concerns about how to effectively use billions of data points that are now being generated by affordable computing power and their connected machines are common. Many SMMs are not sure how to get started and often attribute their caution in adopting AI to one or more of the three common myths about AI implementations in manufacturing:

  1. It’s expensive.
  2. You need a team of AI experts.
  3. You are not ready.

Breaking Down the Myths About AI in Manufacturing

Let’s break down the three common myths:

  1. It’s expensive. Like other innovations before it, AI can be introduced in small ways. An AI implementation can be done on a single machine; a smaller initiative might pay for itself in as little as 45 days. A company with a group of similar machines might be able to pay a flat fee per machine type or family, and roll out the implementation one machine family at a time.
  2. You need a team of IT experts. With many vendors, you do not need any staff expertise. Machine operators may see nothing different in operations beyond an alert to notify them of an issue. You may not need any special training for an implementation. But you do need someone on the business side of your operation who understands the process and how to articulate deliverables for the initiative.
  3. You are not ready.
    “We don’t have enough data to deploy this.” “We have more pressing issues.” “Our equipment is too old.” “We don’t have the resources to manage the project.” These are common concerns, but there are ways to address these issues so you can capitalize on AI technologies. You might want to ask yourself: What is that problem costing you, and how long can you afford to wait to address it? If you don’t put together a plan of action, will you ever be ready to address it?

What a Manufacturing AI Implementation Looks Like

An engagement with an AI vendor usually begins with an assessment to analyze machines and data being collected. The vendor will cleanse existing data and run it through models to see what can be learned. The vendor may recommend adding sensors to fill any gaps in data. Older equipment may not have adequate sensors, but many affordable aftermarket sensors exist.

The vendor will provide an appropriate algorithm — in essence what to look for and how to respond. An outcome could range from an alert to notify machine operators about an upcoming issue or a system shutdown.

A successful AI implementation requires three things:

  1. A specific and measurable problem to solve with demonstrated ROI
  2. The data (or access to data) related to the problem
  3. Executive and cultural buy-in

A manufacturer’s operations and IT teams must work closely together with a vendor in order to optimize the implementation. Projects start small, with an early win, and then progress through the plant floor. It might begin with a single operational problem, and scale to the entire enterprise.

AI in Manufacturing Often Focuses on Avoiding Unplanned Downtime

When an AI vendor cleanses data, it might see spikes in certain activities or outcomes during machine startups or other trends or anomalies. AI looks at a series of factors that could be predictive of events (anomalies vs. normal). What is the explanation for any anomaly? Does a production line need to be shut down, or can it run at 80 percent capacity until a replacement part arrives?

In one real-world example, a company was experiencing unplanned equipment downtime on a primary piece of process equipment several times a year. The replacement part was expensive, which made it costly to have available inventory, but it also had a long order lead time. Each failure cost hundreds of thousands of dollars in lost production. After analyzing data, looking at normalities, anomalies and correlations of failure data, AI was able to identify an imminent failure and provide enough warning to order parts and schedule planned downtime for repairs.

Likewise, a different manufacturer was experiencing failure at the final stage of testing for a small number of giant diesel engines, similar to what are used to power cruise ships. It might have taken months to find the source of the problems using traditional analytical methods. Within 45 days, AI was able to predict 80 percent of failures with some false positives. After further training the model, it now works at 100 percent accuracy, alerting operators to shut down testing in time to make needed repairs.

AI Is a Competitive Advantage Now but Adoption Is Quickly Increasing

SMMs who harness this technology early on have a competitive advantage with improved first-pass yields, more consistent quality, improved OEE and better forecasting. But what is now a differentiator will soon be standard operations. AI in manufacturing is predicted to grow by more than 50% annually through 2027. In fact, many manufacturers with fewer than 50 employees are jumping in.

To benefit from the advantages of AI, connect with the experts at your local MEP Center and learn how AI can help you improve your OEE and prevent unplanned downtime and quality issues.

Melissa Steinkuhl

Melissa Steinkuhl is a Regional Vice President for the South Carolina Manufacturing Extension Partnership (SCMEP), part of the MEP National NetworkTM. She has more than 30 years of manufacturing and consulting experience. Her expertise includes operational processes, strategic planning and change management.

Popular Sponsored Recommendations

Voice your opinion!

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