How AI improves manufacturing quality
By Rich McKay | 4 minute read | August 15, 2018
Artificial intelligence that can see and hear defects in manufacturing?
Yes. AI already helps us reduce traffic congestion, makes cities safer, and limits air pollution. And now it can be the eyes and ears of manufacturing quality checks—identifying wonky engines, missing circuit board components, and scratched screens to remove defective parts and products before they hit the marketplace. Almost any manufactured part could benefit from AI’s gaze.
“Any industry where manufacturing flaws can be detected visually is suitable for AI,” said Jiani Zhang, Watson IoT Director of Product Management.
More than half of quality checks involve visual confirmation. Often manufacturers rely on highly skilled manual inspections, but these can be time-consuming, occasionally dangerous, and costly.
And even when manufacturers use cameras, they are only capturing a fraction of the potential data. Manufacturers use cameras for quality control, but “all that tells you is if a product is a ‘pass’ or ‘not pass,’” Zhang said. “We can do so much more with those photos. What kind of defect was it? Does it need to be scrapped or can it be reworked?”
By applying AI to visual inspections, organizations can identify defects by matching patterns to images that were previously analyzed and classified. And using 3D techniques, they can even classify the severity of a blemish.
Manufacturers also use sounds—or acoustics—in determining quality. Sensors on equipment can detect changes in vibration or sound that may indicate a defect. Acoustic inspections are a huge help for products or equipment that can be challenging to manually inspect. The misaligned jets of a dishwasher make a distinctive sound, as does the sound of a faulty engine.
And AI can learn quickly. It can be trained to recognize both good and bad parts with just a few hundred defect images or sound files. And it will get smarter and improve its accuracy over time as it’s exposed to more defects.
The cost of poor quality in manufacturing
For many manufacturers, quality costs as high as 15 to 20 percent of sales revenue are routine, chiseling away at the bottom line, tarnishing their brand, and frustrating customers. Even thriving companies that strive to deliver the highest quality during every stage of their manufacturing processes are not immune to quality issues that lead to high rates of scrap.
“Manufacturing defects are a huge issue for the industry. In some cases, 50 percent of production can end up as scrap because of defects, while in some complex manufacturing lines the rate of scrap can be as high as 90 percent,” said Odd Myklebust, the IFACOM coordinator at the Norwegian University of Science and Technology in Trondheim.
The Industry 4.0 data deluge
Industry 4.0 applications enabled by IoT are expected to create a new surge of factory productivity, creating value up to USD 3.7 trillion per year in 2025. That increase in manufacturing productivity could be eaten away by quality issues.
Finding new insights will be key. “Manufacturers are sitting on a goldmine of data,” said Zhang. “We hear from customers that their machines have been spitting out data for decades, but they didn’t know what to do with it.”
With connected sensors, companies can gather real-time data about performance and quality to avert possible issues. “IoT devices can be applied to machinery on a factory floor, capturing data on energy usage, temperature, and output. IoT devices can also be outfitted on checkpoints in the distribution process, where they can keep tabs on parts and products as they are shipped from factory to warehouse and beyond,” said Ari Zoldon, CEO Quantum Media.
And with all that IoT data being collected, manufacturing companies need a better way to analyze and see patterns to improve quality. That’s where AI—and its ability to process mountains of data—comes in.
AI can “identify causal problems that led to quality problems”—analyzing data from raw materials, production lines, finished products, maintenance records, and customer complaints.
Quality in the field
Manufacturers can also glean insights from their increasingly connected products after the sale. “IoT sensors embedded within these devices and machines provide hard evidence around actual product operations, responses, and user interaction patterns,” said Bruce Anderson, Electronics Industry Global Managing Director at IBM.
By monitoring how customers actually use their products, companies can potentially alert customer service, warranty management, and even product design to shortcomings in product performance or unanticipated usage. These combined insights pave the way for designing new innovative capabilities and products.
“Sensor data can also record how many times a refrigerator door is opened, or how often certain buttons are pressed,” said Anderson “In the manufacturing process, companies can redesign components or sub-systems based on actual product usage and malfunction data. This will help reduce warranty and maintenance costs, with higher quality products and more satisfied customers.”