The OV20i is a powerful all-in-one artificial intelligence based inspection system. To get the most out of your device we recommend reading the entire manual, however you can start by seeing how example workflows work.
Recipe
A recipe can generally be described as an inspection task. You can create as many recipes as you want to inspect different parts, or different sides of the same part. Each recipe has it’s own imaging setup, alignment block, ROI block, classification block and I/O block.
You can only have one recipe active at a time. You need to activate a recipe before you can edit it. Note: while editing a recipe, it will no longer be able to perform inspection.
Getting a good image is key to a good inspection, spend some time choosing the right lighting, angle, distance, focus, exposure and gain.
Choose the trigger type that best suits your application: manual or hardware trigger.
Aligning your part to a template image allows you to do more high precision inspections on key regions of interest. Having the right imaging makes the aligner easy and accurate. However common pitfalls can make the aligner fickle and unreliable.
ROI Block Here, we break up the inspection task into inspection types and inspection regions. Each inspection type can have as many inspection regions as you would like. So far example if you want to inspect 10 screws and 5 holes. You would create 2 inspection types and name them: “screws” and “bolts”. You would then create 10 screw inspection regions: “screw1”, “screw2”, “screw3”, etc. Finally create 5 bolt inspection regions and name them: “bolt1”, “bolt2”, “bolt3”, etc.
In this block, we build the AI model for each inspection type. First you want to define the potential classes of your model, the default is “good” and “bad”, but this could also be “dent”, “scratch”, “hole”, “texture issue”, “pass”. You can define as many classes as you would like. The caveat is that you need 20 images per class minimum for a model to be effective. Each inspection type has its own set of classes.
Setting up the right I/O for your inspection is usually highly customized integration. We use Node-RED to make this easy. With Node-RED you can setup discrete I/O or use industrial communication protocols.
You can perform actions based on the prediction of the class, the confidence of the prediction, alignment not being found, alignment match angle, and more.