Engineering Expertise

An industrial vision system uses cameras, lighting, optics, and image processing to inspect, measure, and guide production processes at speeds manual inspection cannot match. Integrating one successfully requires disciplined engineering at every layer: hardware selection, lighting design, algorithm development, and controls integration. Each layer independently determines whether the system performs reliably in production.

This article is part of the series on custom machine design and automation and connects to our articles on robotics integration, , PLC programming and industrial controls, and the machine design process.

Watch: Bravo Team engineers discuss vision integration and machine system design in practice. See the full conversation on YouTube.

What Is an Industrial Vision System and What Can It Do?

Industrial vision systems handle a wide range of inspection and guidance tasks in manufacturing. Common applications include:

• Dimensional measurement: verifying part dimensions to tolerances as tight as ±2 microns with calibrated area scan cameras
• Defect detection: surface scratches, voids, cracks, contamination, and assembly errors
• Barcode and OCR reading: 1D/2D codes, alphanumeric characters, and data matrix codes at production line speeds
• Assembly verification: confirming presence, position, and orientation of multiple components simultaneously
• Robot guidance: providing real-time position and orientation corrections to robot controllers for vision-guided pick-and-place and assembly

Vision integration is a core capability within Bravo Team’s machine and system design services. Hardware selection, lighting design, algorithm development, and controls integration are all handled in-house.

How Do You Select the Right Industrial Camera for Your Application?

Camera selection is driven by the inspection task. The four primary parameters are:

• Resolution: how much detail the sensor can resolve; a 5MP camera provides roughly 3x the spatial resolution of a 1.6MP camera at the same field of view
• Frame rate: images per second the camera can acquire; standard industrial cameras range from 30fps to over 500fps for high-speed inspection
• Sensor size: determines the field of view at a given working distance; larger sensors require larger optics but allow wider fields of view at higher resolution
• Interface: GigE Vision (standard, up to 10m cable), USB3 Vision (short runs, high bandwidth), Camera Link (high-speed, short runs), CoaXPress (very high speed over coax)

When should you use an area scan camera?

Area scan cameras capture a full two-dimensional image and are the standard choice for inspecting discrete, stationary or slow-moving parts. Bravo Team works with area scan cameras from Cognex (In-Sight and Dataman series), FLIR Teledyne, IMPERX, and Teledyne DALSA. Cognex In-Sight smart cameras are well-suited to self-contained inspection applications where an integrated lighting and processing solution simplifies deployment. FLIR Teledyne and IMPERX serve applications requiring very high resolution or specialized sensor characteristics.

Why Is Lighting Design the Most Critical Variable in Machine Vision?

Lighting is the most underestimated variable in vision integration and the most common root cause of system failures in production. The goal is not general illumination. It is maximum contrast between the features of interest and the background, so the algorithm has the clearest possible input.

The main lighting techniques and when each applies:

• Ring illumination: even, diffuse front lighting; best for low-relief features on matte surfaces; not suitable for specular or highly reflective parts
• Backlighting: places a uniform light source behind the part, imaging it in silhouette; most reliable approach for edge and hole dimensional measurement; eliminates surface texture noise
• Coaxial (on-axis) illumination: directs light along the optical axis via a beamsplitter; essential for specular surfaces such as polished metal, glass, and PCBs, where conventional lighting creates glare
• Structured light: projects a known pattern onto the part; deformation of the pattern is used to reconstruct 3D surface geometry; used for height measurement, warp detection, and surface reconstruction
• Dark field illumination: low-angle side lighting that makes surface scratches, raised features, and fine texture visible against a dark background; common for scratch detection on flat surfaces

Lighting also requires environmental stability. Ambient light fluctuations from windows or overhead fixtures can produce image variations that overwhelm the algorithm. Shielding the inspection station from ambient light and using LEDs with driver circuits that maintain constant output over temperature are engineering requirements. Our research and development team validates lighting designs under production-representative conditions during proof-of-concept testing.

When Should You Use AI-Based Vision Instead of Classical Machine Vision?

The choice between classical algorithms and AI-based inspection depends on the inspection problem:

• Use classical machine vision (Cognex VisionPro, OpenCV) when: the defect or feature is geometrically definable; the inspection task is dimensional measurement, edge detection, or barcode reading; training data is limited; validation against defined accept/reject criteria is required for regulatory compliance
• Use AI-based inspection (YOLO, Darknet, custom neural networks) when: defect appearance is variable and cannot be captured by geometric rules; multiple features must be evaluated simultaneously; the boundary between acceptable and unacceptable is complex; sufficient labeled training data from production is available (typically 500-2,000+ labeled images per class minimum)

How Do Vision Systems Integrate with Robots, PLCs, and Motion Controllers?

Vision-guided robotic integration requires three elements to work correctly together:

1. Shared coordinate frame calibration. The vision system and robot controller must be calibrated so that a position measured by the camera translates accurately into a robot motion command. This calibration must be verified periodically and re-established if either the camera or robot mounting changes.
2. Latency management. The communication latency between vision processing and motion command execution must be accounted for. If the robot begins moving before the vision result is fully transmitted, the position correction arrives too late.
3. PLC handshaking. The robot must only move when a current, valid vision result is available. Fault conditions in the vision system must produce defined, safe behavior at the PLC level, not undefined robot motion.

For the PLC communication architecture that supports this integration, see our article on PLC programming and industrial controls. For the robot side of vision-guided integration, see robotics integration for manufacturers.

Why Do Industrial Vision Integration Projects Fail in Production?

The four most common root causes:

• Poor lighting design not validated in production conditions. Lighting evaluated in a controlled lab fails when ambient light, part temperature, or machine vibration introduces image variation that the algorithm was not designed to handle.
• Insufficient AI training data. Models trained on small or unrepresentative datasets perform well in development and fail in production. Minimum viable training sets for reliable defect detection typically require 500 to 2,000 labeled images per defect class collected from actual production output.
• No statistical production validation before go-live. Validation against a small sample of known-good and known-bad parts in a lab does not predict production false-positive and false-negative rates. A statistically meaningful sample of actual production output is required before any automated inspection system is released.
• Vision was added as an afterthought. Camera mounting, lighting enclosure, part presentation, and controls interface designed into a cell from the start always outperform systems where vision was retrofitted around a completed mechanical design. As outlined in our custom machine design guide, vision needs to be part of the system architecture from requirements through commissioning.

Want to see Bravo Team’s approach to machine vision and automation? Watch our engineers on YouTube.

Frequently Asked Questions About Industrial Vision Systems

What is the difference between machine vision and computer vision?

Machine vision refers specifically to the use of cameras and image processing in industrial automation: inspection, measurement, guidance, and identification in a production environment. Computer vision is the broader field that enables machines to interpret visual information, spanning consumer software, research, and autonomous vehicles. Industrial vision systems are a subset of computer vision engineered for the deterministic performance, production-speed processing, and automation system integration that manufacturing requires.

How accurate can a machine vision system be for dimensional measurement?

With calibrated area scan cameras, precision optics, and controlled lighting, industrial vision systems can achieve dimensional measurement accuracy of ±2 to ±5 microns on features within the camera’s field of view. Accuracy degrades with field-of-view size, lighting variation, and calibration quality. For sub-micron measurements, laser line profilometers and structured-light 3D sensors are more appropriate than area-scan cameras.

What vision platforms and software does Bravo Team use?

Bravo Team works with vision hardware from Cognex, FLIR Teledyne, IMPERX, Teledyne DALSA, Hermary, and Motorola Solutions, and develops vision algorithms using Cognex VisionPro, OpenCV, YOLO, Darknet, and custom neural network architectures. Platform and software selection are driven by the inspection requirement. See our machine and system design capabilities for more details.

How does a vision system communicate with a PLC?

Vision systems communicate with PLCs via discrete I/O (for simple pass/fail signals), or network protocols including EtherCAT, Ethernet/IP, and Modbus TCP for applications requiring position data, measurement results, or classification codes. The interface must include well-defined handshaking so that the PLC acts only on current, valid vision results. For more on controls integration architecture, see our article on PLC programming and industrial controls.

Ready to Evaluate a Vision Integration for Your Application?

An industrial vision system that performs reliably in production is the result of disciplined engineering at every layer: camera selection, lighting design, algorithm development, system integration, and production validation. The camera is the most visible component. Lighting and algorithm design determine whether the system actually works.

Bravo Team’s vision integration capabilities are fully in-house as part of our machine and system design services. Explore our Engineering as a Service model or contact us to start the conversation.