Machine Learning and AI Improve the Bottom Line

Before the introduction of smart components to industry, manufacturing and processing operations lived in a break/fix mechanical world where maintenance personnel could practically look at a machine and know what was causing it to stop. In those days, long runs of the same product led to fewer changeovers with less tinkering and adjusting.

Today, changeovers occur frequently, and machines are complex and difficult to analyze. Fewer personnel possess exceptional diagnostic insights. With today’s high costs and liability issues, downtime and poor quality need to approach zero incidents.

Frank Latino, product management, electric automation, Festo North America.

Smart components allowed industry to move beyond break/fix to preventative maintenance and predictive analytics using machine learning and artificial intelligence (AI). These technologies take predictive analytics to new levels of performance through early identification of out-of-spec operation and communication to team members, either through dashboards or in easy-to-understand messaging that convey vital, on-point information. Receiving a solution prior to a shutdown means higher uptime, greater quality, and overall improved throughout. Targeted maintenance is an efficient use of maintenance resources.

As advanced as they seem, machine learning and AI are not bleeding edge solutions, but practical tools available now. Festo has found that machine learning/AI can improve process transparency by 100%, lower waste by more than 50%, and decrease product rejection costs by more than 45%. Machine availability can be improved by more than 25%. Unplanned downtime can fall by more than 20%.

AI provided early detection of wafer defects in a wafer sawing process. The AI solution lowered annual waste per machine by $100,000.

Machine learning systems are trained by operational data on components, machines, products, and energy systems. The AI system creates a map that represents the region of healthy performance. Anomalies (outliers) on the map prompt corrective action. As a result, machine learning and AI effectively map the big three operational concerns: health of the process, quality of products and energy consumed. Machine learning and AI systems should be vendor agnostic, not limited to or focused on a brand of component or type, but simply acquiring and analyzing pertinent resident data. Presentation of actionable information should be configured for the end user’s workflow. The end user decides where, when, and how to display information. Machine learning and AI use basic operational data to learn the healthy state of the process, part, or energy system. This data can include speed, distance, pressure, flow, current, temperature, vibration, environmental factors such as humidity, torque and number of cycles in a component’s lifecycle, etc. Key data on product conformance to specification and energy system efficiency provide an equally clear picture of real-time actualities. AI builds a digital bridge between this operations technology (OT) and information technology (IT).

Identification of anomalies, now what?

Early in the machine learning/AI development process, it became clear that end users wanted the AI healthy/not healthy data available in the way most convenient for them, whether on their own dashboards, resident on premises, on edge devices, or all three. Some customers wanted the data integrated with their maintenance management systems. Others wanted alerts sent to mobile devices. Many end users asked whether they could have easily understood messages identifying what the problem was, where it was and what corrective action should be taken. Some of these end users wanted information delivered via text message.

Given these varied requirements, a machine learning/AI system must be flexible enough to enable connections to internal maintenance management software or spare parts management system to create an integrated end-to-end solution. The point for the end user is to ensure the machine learning/AI vendor provides a system customizable to the needs of the operation.

Frank Latino is product manager, electric automation, Festo North America.