Without a manufacturer product number, up to 50 attributes and features may be relevant for the unique identification of a complex replacement part. Simple replacement parts have five to ten features.
Without a manufacturer product number, up to 50 attributes and features may be relevant for the unique identification of a complex replacement part. Simple replacement parts have five to ten features. Taking a ball valve as example, there are about 2500 different variants with several hundred different dimensions. In the end, we can estimate something like 250,000 different order numbers for ball valves.
One image. If the short and long descriptions are missing, then the image fills in for part of the professional description. An example: Male or female thread.
360-degree images replace even more information. Example: Ball valve from various sides
3D renderings help to describe specific features and shapes. In some cases, shapes and silhouettes cannot be described with text.
Depending on the complexity of the part, the options are significantly reduced. Based on clear lists of an assembly or a parts list for a machine, decisions can be made on the basis of all features even without 100% certainty.
Images enable the elimination of a part from consideration within a second.
Out of list of 15 to 20 options, people immediately see the candidates for elimination because the shapes, materials and colors can differ considerably.
Example: Plastic ball valve
In an uncertain situation, images of the replacement parts serve as confirmation.
An artificial intelligence can make a preselection based on image information and search through an image database. Similar to an expert who is only shown images of products.
Based on the image information, an artificial intelligence can extract visible features in a very unstructured, semantic way and use this information to search a text database.
On the basis of image information and text information, an artificial intelligence can search a database containing texts and images in a similar way to how a person would, making use of the best of both worlds to optimize the search results.
An artificial intelligence can propose relevant filters based on image information.
Dimensions cannot be discerned in detail. Only relations. Example image of a port.
Functions and features that are not visible may be extremely relevant, and errors are catastrophic. Example: Pressure and temperature ranges.
Especially in the area of fire protection, important certifications and trade association requirements exist. Simple replacement parts are turned into complex identification cases by decals and stickers.
Pure 3D renderings depend on the starting material and are frequently not informative enough.
Image information is extremely helpful
Visual and textual information about replacement parts should be combined because images alone are unfortunately insufficient.
Neural networks trained on replacement parts are the entry point for an AI-based replacement parts search with images.
In this article, you will discover more information about how many or which images are needed for AI-based replacement parts identification.