Automated visual inspection: more than just data

Features - Features

Leveraging Big Data collected from automated inspection systems allows automotive manufacturers to track quality, rapidly introduce products, and improve supply chain management.

Subscribe
April 7, 2021

Kitov.ai provides AI-based, automated smart visual inspection systems to improve quality, performance, and overall efficiency.
Photo credit: Kitov.ai

Faced with a competitive climate and dynamic consumer behaviors, automotive manufacturers have transformed operations into flexible, smarter environments, allowing for flexible and improved production to meet demand. A key element of this environment – automated visual inspection. Automotive manufacturers can find all critical defects on the production floor and use those data to gather further insights into the causes of deficiencies and make adjustments to improve quality and performance. As systems grow smarter and easier to deploy, capabilities are being pushed even further and manufacturers are implementing deep-learning image analysis in new ways.

Robust data, AI

Automated visual inspection can now detect and measure more variables and components, making it increasingly popular for today’s vehicles which are equipped with a growing number of electronic control units (ECUs), sensors, and software that must be inspected for critical defects and meet strict industry standards. Automated inspection systems provide a vastly more powerful approach and allow a more holistic picture to be seen, due in large part to increasing capabilities of software systems and artificial intelligence (AI).

“With deep learning and deep neural networks, we now have the power to manipulate this Big Data and make significant improvements,” says Corey Merchant, vice president of the Americas at Kitov.ai. “You can capture images and save them for historical records, and then also teach the system what is acceptable to identify.”

Users train software systems to learn the difference between a good image and a bad image. The software then analyzes images and distinguishes relationships between features to create a weighted table or neural network that defines what makes a good or bad part. Manufacturers can analyze data to determine trends, make any adjustments, and drive predictability to maintain quality.

A bonus – the ability to read barcodes and labels.

“This has made the systems much more influential,” Merchant adds. “You’re not only seeing the product, you’re getting a code off that product and can execute serialization and traceability. Users can track that part and identify where it was inspected, or where it was added in the supply chain.”

The integration of AI, motion, and inspection is also transforming the process by permitting more reliability and efficiency.

AI enables systems with robotics and vision to identify parts and learn how to look for them. For example, the deep learning network can capture thousands of images of a screw and know where it’s located so it knows what to look for in the future.

Software packages are now programmed with more semantic terms as well, so the user simply selects buttons labeled with each part or surface area and the system knows what to look for and how to read and locate the part. For parts such as hose clamps, electrical connectors, and harnesses, the system can identify that they’re present and properly positioned before being moved to the next stage of the assembly process.

“Users are always going to want more power,” Merchant adds. “And these trends will continue to expand. Part of my mission is to take out the frustration for end users who may have custom vision and analytical systems sitting idle because they didn’t work as planned. Or maybe they work from day one, but now the system has changed and the solution doesn’t keep up. AI and deep learning are a big crutch and helping to do that.”

Hybrid vision solution

One challenge is that quality control can bring significant expenses to auto manufacturers and the supply chain. However, fully automated systems that improve over time and adjust to the environment can reduce expenses and increase throughput for a promising return on investment (ROI).

Kitov One, developed as a tool to incorporate these elements, is a fully automated hybrid approach combining 2D, 3D, and deep learning technologies to quickly, accurately, and reliably enable industrial applications previously considered too complex for automated inspection.

The 3D vision system finds critical defects, such as bent pins on connectors and ports on electronics, surface damage on metal parts, coverage defects typical to paint and galvanized surfaces, as well as missing components of complex assemblies and sub-assemblies. Using pre-set algorithms, the software computes and controls image acquisition and processing.

“Deep learning requires many stored images to be effective which takes time to acquire when in production,” Merchant says. “The conventional machine vision tools enable excellent performance of the system from day one. Eventually, the AI/DL tools will kick in using actual acquired images from the customers’ assembly lines which will make the Kitov.ai vision system even better over time.”

Merchant goes on to explain that the system can adapt to any changes that occur, such as those in production or in the environment. This can cause temporary dips in system accuracy as the system learns these new variables, but the conventional machine vision tools enable the system to maintain high levels of accuracy and results are not affected by environmental changes. Another production benefit is the ability to go back to historical data and inspection reports to perform regression testing. This allows them to add additional inspection test criteria after the fact to determine when an unanticipated defect started to appear and track its progress over time.

“The Kitov.ai solution allows end users to program or modify inspection plans without previous vision or robotic programming experience,” Merchant explains. “Our software uses semantic terms in the user interface such as screw, label, barcode, surface, and the AI algorithms then set the optimal inspection plan, including setting the ideal robot path.”

This simplified language is important for leveraging AI and continuing to boost quality and flexibility in production. In the future, Merchant expects users to increasingly adapt their own terminology and classifications. This will allow for a more useful inspection tool and will help overcome the challenges of complexity, ultimately creating that optimized test plant that players in the field seek.

Kitov Inc. USA https://www.kitov.ai

About the author: Michelle Jacobson is the assistant editor of TMV. She can be reached at mjacobson@gie.net or 216.393.0323.

OES

High precision vertical elevator stages

The AT10-60-01, AT10-60-02, and AT10-60-03 motorized elevator stages feature 120mm x 80mm (4.724" x 3.149") high stability tables and 60mm ±2mm (2.362" ±0.078") of vertical travel with 5.0kg (11.0 lb) load capacity. The scissor lift design combined with precision roller bearings and slide rails ensures smooth motion, high parallelism while raising and lowering the table.

Z-axis stages are suitable for: microscopy, inspection, metrology, positioning, photonics, sampling, and laser drilling and machining.

Each table has threaded holes for fixtures or tooling, and countersunk holes in the base allow easy integration into new or existing applications.

Each stage requires a motion controller ordered as a complete plug-and-play system with RS-232, USB, joystick, trackball, and keypad control.

OES (Optimal-Engineering Systems) http://www.oesincorp.com

Wintriss Controls Group

Stamping sensors

To protect stamping and forming press dies, an updated sensor family includes broader-range proximity sensors, short and long range diffuse reflective sensors, a distance-settable optical triangulation sensor, and an ultra-long infrared through-beam sensor.

  • Proximity sensors have 4x the range of similarly sized inductive proximity sensors, in addition to proximity sensors that detect only non-ferrous metals, suitable for detecting aluminum or copper parts from tooling
  • Long range diffuse reflective sensors use infrared instead of visible light, minimizing false actuations caused by oil mist in the sensing field
  • Special distance-settable optical triangulation sensor suitable for challenging applications where a part must be detected against objects such as a chute
  • Ultra-long-range infrared through-beam sensor excels in lubrication intensive environments, enabling reliable, repeatable operation

All sensors are IP-67 rated and can withstand harsh in-die environments.

Wintriss Controls Group https://www.wintriss.com

Capture 3D

Automated scanning cobot

The GOM ScanCobot is a mobile automated 3D scanning system that integrates with GOM’s blue light 3D scanners ATOS Q and ATOS Core. Users can create efficient automation routines that produce precise, accurate, and repeatable measurements.

The system includes a motorized rotation table, a PC for 3D data processing, and intelligent software for 3D scanning, inspection, and automation.

It’s equipped with GOM software for parametric dimensional inspection and includes the virtual measuring room (VMR) add-on and the Kiosk Interface for quick offline and online programming. Software accelerates measurement by automatically and simultaneously performing measurement planning, digitization, and inspection. The robot and software work together to find measurement positions while learning optimal movements, creating an easy-to-operate intelligent, guided user interface on a single platform.

Capture 3D Inc. https://www.capture3d.com