How To Automate Defect Detection with Machine Vision

Quality assurance (QA) began as a philosophy during the Industrial Revolution. In the late 19th century, American engineer Frederick Winslow Taylor developed a new system for improving efficiency and productivity in factories. His methods focused on providing training for employees, rather than allowing them to just learn on the job. The protocols Taylor implemented were also based in science, which is why he also promoted rigorous documentation.

In 1911, he published a book explaining his QA methodology. Taylor’s techniques have since proved integral to the world of manufacturing. Automation has largely replaced the manual QA methods he promoted, as technology and software have both advanced to perform increasingly complex tasks. As an important aspect of the QA process, defect detection has evolved as well, with machine vision now becoming an important means for ensuring efficient production.

The Importance of Quality Assurance

Quality AssuranceBefore looking into the custom automation that’s now being used for detecting defects, it’s important to understand the importance of QA in modern manufacturing. Automation has certainly enhanced the quality of goods coming out of factories, but defect detection has been a part of manufacturing prior to the introduction of this new technology.

Developed in the 1950s, a system known as the Failure Mode and Effects Analysis (FMEA) became one of the first structures for improving products and processes. The FMEA methodology is still used today, though variations have been developed for specific purposes. It helps manufacturers anticipate failures, either in the design stage or the manufacturing process.

The steps for developing an FMEA are:

  • Define scope of FMEA: This step involves collecting key information in order to determine what aspects of a product cause failure. This step involves investigating documented cases of defects in order to develop a plan for identifying and reducing their prevalence in end products.
  • Strategy development: This stage of the process looks at the information from these documented events, calculating what functions are likely to fail, how they’re likely to fail, the effects of this failure and how to measure the severity of a failure.
  • Defining processes: This point in the development stage involves looking at the potential causes of failures, which are selected either from design inputs or past failures.
  • Determining mode of failure: This part of the development process looks at adding a means for detecting defects by ranking them, which helps ensure that a product is properly designed and what may happen should specific defects in a product are passed on to the end user.
  • Calculating risk and prioritizing remediation: Actions taken in the three previous steps – strategy development, defining processes and determining the mode of failure – are then assigned a risk priority number (RPN) to determine how these issues should be remediated. To calculate the RPN, the severity and occurrence of a potential failure is considered, along with detection rankings.
  • Taking corrective action: This element of the process consists of countermeasures taken as a means to reduce risk of failure.
  • Assessment: Once actions to mitigate risk have been taken, the ranking system adjusts to calculate a new RPN to improve product design or the production process.

This manual process for defect detection is increasingly being enhanced via automation. Engineering defects can be spotted much more easily by technologies like machine vision, whereas human inspectors are inconsistent in their classification of defects. This inconsistency leads to variable quality of products depending who is doing the inspecting.

Using Automated Visual Inspection for QA

Often referred to as machine vision, these programmable systems offer a means for inspecting an array of industrial applications. They utilize a smart camera specifically calibrated to automatically inspect products and processes in real-time. They essentially replace human inspectors, helping to identify any irregularities along the production line or problems with the manufacturing process.

Machine vision inspection systems increase efficiency, and generally pay for themselves within 24 months. While they perform tasks more proficiently than human eyes, they do have their limitations. Many of these are due, however, to the fact that designers often don’t take into account

Machine vision systems sometimes have difficulties due to factors like:

  • Camera resolution
  • Lack of quality data
  • Lighting anomalies
  • Shutter speed
  • Type of defect

To detect defects, machine vision systems look at a product’s surfaces and dimensions during manufacturing. Automation of QA is very much based upon statistics, much like the manual FMEA models. Computers are very good at forecasting probabilities and can be programmed to detect whether a product on the assembly line falls within specific parameters. From these calculations, artificial intelligence (AI) software then helps the system determine whether a product is defective.

This doesn’t mean the end of human QA inspectors, however. Machine vision capabilities depend on cataloguing defects so that the system can automatically do its work. These directories, collated with the help of human QA inspectors, help the AI ascertain what’s acceptable and what isn’t.

Types of Machine Vision Systems

There isn’t just one kind of machine vision system used for manufacturing. Automation of this type is diverse and can be used for numerous applications. Factors that affect the type of machine vision system needed include a manufacturer’s budget, properties of the process or product, inspection speed needed and other factors.

The different types of machine vision systems include: 

  • Custom automation packaging: These systems specifically look at packaging, designed to both meet specific quality standards and limit false rejections; they inspect for such things damage to items, misapplied labels and any deformities to a product packaging’s molded features.
  • Defect detection: These systems catch and resolve issues related to the product or its packaging; working in real-time, they identify broken components, faulty seals, misplaced labels, surface deformities and other defects.
  • Print and code inspection: By identifying damaged or incorrect product labels, these systems check for things like unreadable barcodes, text quality, incorrect nutritional or ingredient labels, artwork quality and other labeling issues.
  • Process control: These machine vision systems work to perfect processes on a production line, enabling precision control of robots, gathering and applying measurements within data logging systems, collecting historical process data, logging data trends, monitoring systems and other imaging tasks along the production line.
  • Tolerance measurements: When a machine vision system is integrated into a production line, it helps ensure the quality of products, making these production standards repeatable; these systems can verify component tolerances, compare finished products with computer-assisted design (CAD) images, count threads, inspect heat seals, scan with cameras or lasers, analyze radiometric data and provide other information regarding product fabrication.
  • Webbing inspection: These systems use real-time image processing to detect any damage to webbing, while also resolving these defects by removing damaged webbing sections.

Manufacturing Automation for Defect Detection

Automated visual inspections offer a means to improve the quality of end products while making manufacturing processes more efficient. Advances in AI have made this possible, with machine vision now capable of performing better in certain cases than human eyes. Additionally, they’re ideal defect detectors as, unlike humans, they don’t get distracted, tired or sick.

Developing a custom automation system for visual inspections involves these steps: 

  • Camera installation requires first choosing the best camera and lens for the application, along with ensuring the camera is well-positioned and has optimal lighting.
  • The AI needs images to be able to accurately perform visual inspections, so there must be adequate storage for these, with backups performed regularly.
  • Each image should have annotations to assist the AI when doing inspections, with which machine learning algorithms should also be developed.
  • Using AVI video files, a highly accurate model should be fabricated to train the AI and assist in its deployment.
  • These models should then be validated so that the AI can properly conduct its tasks, which entails creating a document to detail what’s in the AVI system.
  • As a final step, a dashboard should be built so that operators, using process control software, can check on the inspection results; this dashboard can also include such things as the proportion of defective products over time, number of defects by category and other QA statistics.

AI has enabled manufacturers to perform tasks that once required human brainpower with the support of machine vision. It has enabled this technology to be used for a wide range of applications, combining hardware with software to facilitate QA manufacturing automation. Engineering these automated visual inspection systems to perform ever more complex QA duties is the future of machine vision technology.