Smart cameras & vision sensors

How to choose a smart camera or vision sensor

A smart camera puts the sensor, the processor, the inspection software, the I/O and the network connection in one housing, and most models add the lens and the light. It is set up once from a PC, then runs on its own and sends a pass / fail or a measured value straight to the PLC. Choose a smart camera when one or two views of a part answer the question; choose a PC-based system with area scan cameras when you need many cameras, very high resolution or custom code.

1. Start from the inspection task

List what the camera must decide, then check the tool set of the model. Typical tools here are presence / absence (edge, spot, contour and pattern existence), counting, measurement (width, diameter, point-to-line distance, angle), pattern location with a position fixture that makes the other tools follow the part, colour checks on colour models, and code reading and OCR. Some models add deep-learning classification for defects that rules cannot describe. If the main job is reading codes at speed, a fixed code reader is usually the better fit.

2. Resolution and field of view

Most models here have up to 3 MP; 5 MP and 12 MP models cover larger fields or finer detail, and the range goes up to 25 MP. The rule is the same as for any camera: field of view / pixels across = object size of one pixel, and the smallest feature should span at least 3 pixels. Frame rates run from 10 to 120 fps, which is enough for most part-by-part inspection but not for very fast continuous lines.

3. Lens and focus

Integrated M12 lenses are the norm, most often 8, 12 or 16 mm, with shorter and longer focal lengths for wide or narrow views. Check the focus method:

• Manual focus: fine for a fixed working distance.
• Mechanical autofocus: focus set from software, so a recipe change can move it.
• Liquid lens: focuses electronically in milliseconds, for lines that run parts of different heights.
• C-mount: takes any C-mount machine vision lens, including telecentric and macro lenses, at the cost of a larger housing.

Use the lens focal length calculator to check the field of view at your mounting distance.

4. Built-in light, or an external one

The built-in ring of LEDs is white by default; red, blue, IR, UV and two-colour versions are available on many models, and caps with a diffuser or polariser cut glare on shiny parts. A built-in light works at short distances on matt parts. For shiny, curved or transparent parts, switch it off and use an external ring light, dome light or backlight triggered by the camera's output.

5. Connection to the machine

Check the protocols your PLC speaks. Models here support combinations of PROFINET, EtherNet/IP, Modbus, TCP and UDP, FINS, MELSEC/SLMP, RS-232 and FTP for saving images. Count the I/O: 2-3 opto-isolated inputs and 2-3 outputs on a 12-pin or 17-pin M12 connector is typical, for the trigger, the reject output and recipe selection. Recipes (jobs) are stored in the camera - up to 32 on some models - and can be switched from the PLC, serial commands or operator buttons.

6. Environment

Smart cameras run on 24 V DC, are rated IP65 or IP67 on most models when the lens cap is fitted, and work from 0 to 50 °C. Many are tested to IEC 60068 for vibration (10-55 Hz) and shock (30 g), which matters on presses and moving axes.

What this range covers

• Resolution from 0.3 MP to 25 MP, mono and colour, mostly global shutter.
• Integrated optics from about 2 mm to 25 mm focal length, plus C-mount and liquid-lens models.
• Gigabit or Fast Ethernet; a few USB 2.0 and serial models.
• High-end models with 8 GB memory and 64 GB storage that run a flowchart vision platform with deep-learning modules.
• Open x86 smart cameras that come with an image-capture SDK for your own software (no vision software included).

Typical applications

• Presence and position of caps, labels, seals, clips and screws on assembly lines.
• Go / no-go dimension checks such as hole diameter, gap width and part length.
• Reading lot codes and date codes with OCR, and verifying 1D and 2D codes.
• Colour and orientation checks before packing.
• Guiding a robot or pick-and-place unit with the located part position.
• Mould protection: checking the mould is empty before it closes.

Common mistakes

• Expecting a vision sensor to measure to a few microns: a 1.3 MP sensor over 50 mm gives about 0.04 mm per pixel.
• Relying on the built-in light for shiny or curved parts and then fighting reflections.
• Choosing a fixed-focus model for a line that runs parts of different heights.
• Not checking the PLC protocol before ordering.
• Using free-run instead of a hardware trigger from a photoelectric sensor, so the part is in a different place in every image.

Questions

Smart camera or PC-based system?
A smart camera is simpler to install and maintain for one or two inspection points. A PC or vision controller with separate cameras wins when you need several cameras, high resolution, 3D or custom software.

Do I need to program it?
No code is needed for most jobs: tools are set up in the configuration software and results go to the PLC over the chosen protocol. The open x86 models are the exception and need your own software.

Can it read barcodes too?
Yes, models with code tools read 1D and 2D codes as part of the inspection. For high-speed or multi-code reading, a dedicated code reader is faster and more tolerant.