Cognitive inspection: IBM Visual Insights

By | 2 minute read | July 4, 2017

The adoption of automation and IoT technologies within the manufacturing industry has led to speedier production, better asset management and significantly reduced downtime. But inspection processes, which are often fully manual and reliant on staff availability and expertise, can cause something of a bottleneck. Cognitive capabilities can help reduce inspection time and improve consistency in detecting defects, reserving human expertise for when it is truly needed.

IBM Visual Insights: a cognitive inspection solution

Manual inspections within a manufacturing environment can be time-consuming and occasionally dangerous. Here to help, according to this briefing report, is new offering IBM Visual Insights – part of the IBM Watson IoT for Manufacturing product portfolio.

The solution compares super high definition images from the manufacturing floor against a library of images displaying known defects, to detect faults in parts, components, assemblies and products.

How does the solution work?

At the heart of the solution is advanced image recognition and cognitive analysis, together with continuous machine learning. Model managers and data scientists use their combined expertise to put together a library of known defect images, depicting good and not-good parts, for comparison with images captured from the manufacturing floor. A ‘not-good’ image might include missing components on a circuit board, for example, paint bubbles, surface scratches, incorrect labels or corrosion, among many other possibilities.

Any particular product might have dozens of images representing a variety of typical faults, which are classified by the model manager to allow for easy recognition. The images and their classifications are sent to a data scientist, who trains the cognitive model to recognise them using the NVIDIA Deep Learning GPU Training System (DIGITS). Cognitive capabilities analyse the images, comparing good parts with not-good parts, to develop a cognitive model that can be applied to edge computing servers connected to cameras on the factory floor.

Confidence thresholds

If an image captured by the high definition factory floor cameras corresponds to a known defect in the image library, the fault will be flagged for review. Since 100 percent correlation between two items is unlikely, the inspection supervisor can set an acceptable threshold, below which a notification for review will be generated.

For example, if a cognitive model determines that there is a perfect match between an inspection image and a defect image, the confidence level would be 100 percent. If the inspection image is not an exact match, but strikingly similar, the level might be lower, say 85 percent.

This metric allows inspectors to review items with confidence levels that fall below a predetermined threshold, and apply human expertise to identify new types of defect. IBM Visual Insights’s cognitive capabilities enable continuous learning and improvement, as the model takes feedback from human inspectors.

A birds-eye view

Cloud implementation means that the inspection process can be centrally managed through a dashboard, complete with wide-ranging reporting metrics. Reports can be generated in real-time, or at given intervals as needed, and summarise overarching activities as well as drilling down into individual processes.

Learn more and read the briefing report

Read the full report for further details about IBM Visual Insights. To learn more about how IBM is revolutionising the manufacturing process, take a look at our website and speak to a representative today.