A practical automated visual inspection system can significantly reduce the risk of defective products and lower production costs over time.
If you're manufacturing small, precision parts, quality drives the entire manufacturing process. One way to help ensure the quality of such products is to use automated inspection systems known as machine vision systems. Machine vision systems use computer vision technology to automatically inspect parts for many different defects ( such as contamination, scratches, dents, or deformation caused by manufacturing failures ) and specifications ( primarily dimensional irregularities ) . They also collect data that helps improve manufacturing efficiency, geometry, and tolerances. Most importantly, considering the varying labor costs across different regions, automated inspection is less expensive than manual inspection, with a payback period typically of two years or less.
However, companies should pay attention to the following six key factors before implementing automated inspection projects.
I. Each part has its own solution.
Designing automated visual inspection systems to test different part types has proven to be a challenging task, as part geometry may be occluded or shadows may obscure relevant areas. These limitations are often a consequence of product design, as automated inspection is not part of the product design process. Accurate analysis is aided by defect specifications defined by lighting and processing cycle time, illumination, resolution, and camera speed.
If a suspected defect is found, the operator has the opportunity to inspect it again, while the machine does not. In the design process of an Automated Visual Inspection (AVC) system, every potential defect must be considered for each product. Even similar parts present challenges related to specific materials or product designs, thus necessitating a customized approach.
Automated visual inspection systems are complex combinations of cameras, lighting, and processing systems, including some sophisticated image processing techniques.
If a company is exploring the implementation of an automated visual inspection system for quality control, it makes the most business sense to start with the company's highest-volume parts or very similar parts (such as O-rings). O -ring inspection systems can inspect a wide variety of different components. Configuring a system to test different types of O -rings is relatively easy because O -rings are very similar to each other and have simple geometries ; therefore, a single system can inspect different part types.
II. Dimensional Measurement and Surface Inspection
Generally, automated vision inspection systems in factories fall into two categories: the first type is primarily used for dimensional measurement, and the second type is used for surface defect detection. Of these two, dimensional measurement systems are the easiest to develop. Furthermore, statistical methods for calculating solution capabilities (e.g., measurement system analysis) are readily available for machine design because all aspects of the automated vision inspection system can be estimated in advance. By clearly defining part specifications, the system can identify key parameters, even for parts with the most stringent dimensional tolerances. Once these key parameters are confirmed, the inspection system can begin to be implemented correctly, with computational methods used to validate relevant parameters such as system lighting, speed, or camera resolution.
On the other hand, surface inspection can be very challenging. Documentation instructing human quality inspectors on what defects and dimensions to check when manually inspecting parts has contributed to the development of automated vision inspection systems.
However, these instructions do not provide sufficient guidance for software engineers developing automated visual inspection systems. While human inspectors may fully understand these instructions, software developers need more information.
For example, a quality directive might declare a specific type of defect as NG (Not Good) . Literally, this would require the inspection system to have infinite resolution. Instead of simply saying NG , developers must provide quantifiable defects so they can fit them into the system design requirements. If very small defects are NG , the system needs to be able to identify them, thus requiring even higher image resolution.
Companies often face two challenges : a defect needs to occur and be noticed before it can be corrected ; communicating the characteristics of a specific defect to a computer is difficult. Therefore, companies typically implement "basic" checks across all areas to at least have a chance of discovering unknown defects. Traditionally, machines are built based on these defects because they are already available. Some defect prediction may be feasible, but this increases the risk of "false" scrap.
Reaching an agreement on the size of a defect can be difficult because many factors must be considered. Quality engineers must collaborate with design engineers and even machine suppliers to arrive at appropriate metrics.
III. Definition and Classification of Defects
Because defining and classifying defects is a crucial part of the development process, quality engineers should create a defect catalog before starting the development of an automated visual inspection system. Defect classification is not only a list of all defects the system must be able to detect, but also a collection of those that are within or close to acceptable limits. This part is so important because the system will perform specialized verification by inspecting these components.
As shown in the spreadsheet, the defect catalog includes the type of defect, its severity, probability, and critical size. All of this information helps software developers prioritize the detection of common defects.
Defect definition and classification is the most critical prerequisite for the successful implementation of an automated visual inspection system project.
For each defect, it's best to have at least two samples, one of which should be an edge defect. Depending on the complexity of the part, defect classification typically involves 60 to 100 parts. Having a sufficient number of samples is generally preferable, as identifying parts that don't add knowledge to the catalog is easier than discovering missing important information.
The biggest challenge in developing a defect catalog classification system is communication between the quality department and the visual inspection system developers. The visual inspection system developers need to understand the production process, while the quality department must understand the limitations of the inspection system.
For example, quality engineers understand some of the terminology used in the manufacturing process. However, they must describe specific characteristics such as shape, flexibility, and thickness in detail to vision developers to ensure the system can identify defective parts rather than reject good ones. Depending on the cause and type of defect, different methods are required to detect them.
IV. System Check Process
In order to reliably find a specific defect, the program must perform five steps:
1. Part positioning: The program must position the part in the image to compensate for minor changes in the processing system.
2. Image Segmentation: The program segments the image into functional regions of interest. Each region is defined by a finite set of defects and a single function relative to the part, such as a sealing lip, outer diameter, top surface, etc.
3. Image Standardization: Removes all content typically found in the "good parts" section from the image. This includes design / functional lines and common product variations. This is a core function because it compares each new image to the history of the "good parts" section and highlights deviations from this history.
4. Signal / Noise Optimization: At this stage, the program filters the image, extracting potential defects from known noise. This step needs to consider specific defect characteristics, such as dark contamination, bright marks, and vertical streamlines.
5. Defect Detection and Classification: Finally, the program identifies suspicious areas and assigns a classification value, which is used for a pass or rejection decision. Additionally, the program records whether the part has any acceptable defects. This data can be used to prevent production problems from occurring before they do.
V. Ensure high-quality images
Good inspection relies on high-quality images. Here are a few things a factory can do to ensure that a vision inspection system produces optimal images. Image quality typically depends on lighting and camera settings, but the type of defect must also be considered. Other factors include reducing the number of false rejections ; staying within the required cycle time ; and making full use of available machine space.
Optimal settings allow for the inspection of minimal relevant defects.
For example, a factory can take images of parts before and after tool cleaning or manufactured with different tool sets to understand how process variations will affect image quality. By analyzing these images and applying their knowledge of the limitations of image processing software, the company will be able to devise a workable solution. Software limitations sometimes mean that a company must develop more complex lighting and camera settings. At other times, hardware limitations increase procedural complexity or development time, or in some cases, mean that you have to endure more rejections than you would like.
VI. Large investment = large return
Developing and installing an automated vision inspection system is always a significant investment for any company, but it's often a worthwhile one. The biggest benefit is reduced costs, as automated inspection is less expensive than manual inspection. Another benefit is the ability to develop custom-designed automated vision inspection systems, which can significantly reduce the risk of performing poor inspections and delivering defective parts. It can also detect anomalies before defects are apparent. This data can alert engineers to problems so they can be corrected before defective parts are produced.
Automated visual inspection systems have reduced the number of customer complaints, with complaint rates potentially dropping to zero for some part types. This trend is likely to continue as "intelligent systems" train themselves to find parts that differ from typical products. Intelligent systems make automated inspection systems even more attractive because they reduce the risk of programming errors, as well as the time and cost of implementing new solutions.
