Quality inspection shouldn’t just happen at the end of the line.
Manufacturers are using machine vision to inspect components throughout production, identifying problems before defective parts move downstream.
Machine vision combines digital cameras, specialized lighting, and software to perform these inspections automatically.
The result is earlier identification of problems, fewer defective parts, and better visibility into what’s happening with the manufacturing process.
In-process machine vision can range from simple 2D inspection systems to advanced 3D and AI-enabled technologies.
Machine vision system types include:
Choosing between 2D and 3D vision systems comes down to what needs to be inspected. 2D vision can be highly effective for visual defects and X-Y measurements, while 3D vision becomes more valuable when depth, height, volume, or complex surface geometry matters to quality control.
Does your application require 2D or 3D machine vision?
Traditional 2D machine vision works by capturing an image of a part and then analyzing its features in the X-Y plane, measuring dimensions, and identifying visual characteristics or defects.
Key machine vision inspection types can be broken down into the following five core functions:
Alternatively, customers may choose to clarify the system’s function by asking these questions:
In electronics manufacturing, for example, 2D vision can be used to confirm that chips or other components are present and properly seated on a printed circuit board (PCB). It can also be used to identify problems like broken traces or solder defects.
Another example is verifying the placement of printed information on a product or package. A 2D vision system can check whether a label, logo, date code, or other printed feature appears in the correct location and flag parts where the print is missing, misaligned, or outside an acceptable tolerance.
For 2D vision, camera resolution should be based on the smallest feature or defect the system needs to identify within the required field of view. MCE can help determine the resolution needed to reliably perform the inspection without paying for capabilities the application doesn’t require.
Lens and lighting selection are also critical to getting a robust image that enhances critical features of the inspection. Different lighting techniques emphasize edges, surfaces, or other characteristics that might otherwise be difficult for the system to detect.
For example, diffuse lighting that illuminates a part evenly from multiple angles can minimize the appearance of scratches, dents, and other surface irregularities when the goal is to inspect printed features. But if a manufacturer needs to find those scratches and dents, dark-field lighting can do the opposite. By directing light at a low angle across the surface, edges and imperfections stand out.
This is why MCE evaluates lighting as part of the complete machine vision application. In our lab, our team can test customer parts using different lighting techniques to determine which approach best highlights the specific features or defects the system needs to inspect.
Some inspections can’t provide the required data using only 2D inspection systems. For instance, depth may be an important measurement for the inspection. A 3D vision system captures depth in addition to the visual information available from a conventional 2D image, providing details about the physical shape and height of an object.
For example, manufacturers inspecting injection-molded plastic parts or decorative wood components may need to verify complex surface profiles. A 2D camera can confirm the presence and location of a feature, while a 3D system can measure its height, depth, shape, and overall geometry. This allows the system to identify areas where the contour or profile of the finished part differs from what is expected.
A useful way to decide whether AI belongs in a machine vision application is to ask how easily you can describe the defect.
If the requirement is “these two features must be 2 millimeters apart,” a rules-based system can typically be effective. But if the requirement is “This surface doesn't look the way a good part should,” AI-based anomaly detection may be better suited to the job.
AI-driven machine vision systems can use the same high-resolution cameras, advanced lighting, and smart analytics to detect defects, verify assemblies, and ensure product conformity.
One example of a defect that can be difficult to define occurs during stamping. Consider a stamping press cutting the same metal disc repeatedly. If a slug gets caught between the die and the material, it can leave a defect on the disc. Because that defect can appear in various locations, shapes, and sizes, it can be difficult to define what the vision system should look for, especially when the part has a visually busy or complex background.
AI can recognize patterns and variations associated with accepted and unacceptable parts. This means a level of inspection quality that was not possible just a few short years ago. It also makes certain inspection problems more practical to automate.
This doesn’t mean that AI should replace traditional machine vision in all areas. It expands the range of inspection capability and automated solutions. Some inspections may continue to be better spotted with traditional methods, while variable surface characteristics or difficult to describe defects can benefit from AI.
In many cases, these approaches can work together in one production line, machine, or even inspection cell.
Related: Automation for Modern Manufacturing
Finding a defect is just one piece of the puzzle. Manufacturers need to know details about the component as it travels through the process.
Machine vision can identify individual parts through barcodes, markings, serial numbers, or other identifying features. Inspection results can then be linked to a specific part or production step. The information can be passed to subsequent machines, processes, or systems such as PLCs, robots, laser markers, and MES systems.
This record helps manufacturers make operational decisions about whether a part should continue through the process or be rejected in real time. This is especially important in highly automated and regulated processes in industries such as automotive, medical devices, food and beverage, and electronics.
Machine vision can tell you a lot about a part, but some quality requirements can’t be visually verified. In those cases, vision may need to work alongside other sensing and automation technologies to provide a more complete picture of the part or process.
Consider an assembly operation where a component must be installed in a specific way. Machine vision can confirm that the component is present and positioned correctly, while force verification, such as load-cell or other torque measurement technologies, can confirm that the appropriate amount of force was used during assembly. In other words, a component can look correctly assembled but still fail to function as intended.
In one MCE project, we used machine vision to verify that keyboard keys were present, properly positioned, and correctly marked. An actuator was then used to verify the physical operation of each key, including its travel and the force required to depress it. Together, the technologies could identify both visual and functional defects.
Machine vision can also trigger what happens after an inspection. If a vision system identifies a defective part moving down a conveyor, for example, it can communicate that result to a robot or actuator that removes the part from the line. Inspection data can also be captured and analyzed to help manufacturers identify patterns or determine where defects may be originating.
This is why you need to look beyond the camera when designing a machine vision application. A complete machine vision solution may include lighting, sensors, controls, communications, actuators, data collection, and other technologies needed to turn an inspection result into an action.
MCE approaches machine vision as part of the larger production system, not as a standalone application. Our team works with customers to understand the inspection problem, the surrounding process, and what needs to happen after a defect is identified.
That may involve cameras, lenses, lighting, sensors, controls, communications, actuators, and data collection. And when the solution extends beyond machine vision, MCE can draw on expertise across its broader network in areas such as compressors, filtration, flow control, fluid power, and other production technologies.
Our support can continue after the system is installed and commissioned. On-site support can be provided and, when remote connectivity is available, MCE engineers can connect to a customer’s vision system to troubleshoot issues, make adjustments, and help get the line back up and running without waiting for an onsite service visit.
Reach out to our industrial controls team to schedule an onsite consultation and review of your production line. We’ll help you determine where machine vision fits and what other technologies may be needed to build a reliable inspection process.
Get your engineered solution. Schedule a consultation with an MCE Industrial Controls representative today.