Manufacturing

Visual inspection for improved quality in manufacturing

Share this post:

The challenges of quality inspection in manufacturing

Manufacturing operations strive to deliver the highest quality during every stage of the production or assembly process. Over half of these quality checks involve visual confirmation to ensure the parts are in the correct locations, have the right shape or color or texture, and are free from any blemishes such as scratches, pinholes, foreign particles, etc. Automating these types of visual quality checks is very difficult because of the volume of inspections, product variety, and the possibility that defects may occur anywhere on the product and could be of any size. This is where a new offering from IBM Visual Insights – delivers its highest value.

Learning from defect images that are ‘OK’ and ‘NG’

Taking advantage of IBM’s experience in Deep Learning used by Watson, IBM has developed a new offering for manufacturing clients to automate visual quality inspections. Images of normal and abnormal products from different stages of production can be submitted to the centralized ‘learning service’ that will build analytical models to discern OK vs NG characteristics of parts, components and products that meet quality specifications (OK) and those that don’t (NG). Further, if there is a need to classify defects into different types to address potential root causes and fix the quality issues, the IBM Visual Insights offering can be trained to perform such tasks with a high level of confidence.

Cognitive for continuous improvement in defect recognition

Based on advanced neural networks, the models trained by IBM Visual Insights can be deployed on pre-configured hardware on the factory floor so that there can be very little decision latency during production. The solution can learn continuously by taking feedback from manual inspectors who can review the automated classification and override them based on human judgment. The corrective information along with the image from the production floor is then included in the next training cycle for that analytical model, thereby improving its ability to discern in the future. Such a Cognitive approach is unique in the industry.

Reduce dependency on manual inspection

The IBM Visual Insights offering delivers reliable results with low escape rates to reduce the dependency on specialized labor and to improve throughput of quality processes across multiple industries. The solution is being tried out successfully by several global corporations producing electronics, automotive, and industrial products. If you have manufacturing inspection needs which could benefit from IBM’s cognitive capabilities please, take a few moments to learn more aboutIBM Visual Insights.

Learn more about employing IoT solutions to drive more up-time and lower costs with this report: Using the Internet of Things for preventive maintenance. Or talk to an IBM expert today about your specific questions.

You can also keep up to date with IoT stories like this, by signing up to the monthly IoT Sense newsletter.

Worldwide Electronics Industry Lead, IBM Cognitive Solutions Team

More Manufacturing stories
By Lauren Longhi on November 9, 2018

The top 5 takeaways from the Industrial IoT World conference

From October 29-30, companies from across industries gathered in Atlanta, GA for the Industrial IoT World conference to accelerate their projects “from inspiration to implementation.” As a leader in the IIoT space, IBM was invited to share its expertise. I joined IBM five months ago, and was excited to see our capabilities in person. Here […]

Continue reading

By Ashok Kumar on October 8, 2018

Are MES systems dead? How IoT and AI are transforming the shop floor

Manufacturing is changing due to advancements in the Internet of Things (IoT) and Artificial Intelligence (AI). With  consumer demand for customization driving new requirements for speed and agility, many manufacturers are struggling to keep up. On top of this, more competition and increased demand is putting higher pressure on manufacturers to increase overall plant performance.  […]

Continue reading

By Chris O'Connor on October 2, 2018

Do you have the needed data around yield, throughput, and quality to make sound operational decisions?

Here is a bold idea – your machines, your connected systems, your factories all diagnosing themselves and predicting needs before they affect production. What do you think when you read that? Are you thinking early and accurate identification of defects, improved production, optimized throughput, less waste and no product recalls? You should. Because that is what […]

Continue reading