Many manufacturing companies have identified a critical manual quality visual inspection check within their manufacturing workflow as a process that is repetitive, time consuming and open to errors. In the aerospace sector, for example, every part that is manufactured is machined in pairs, left and right. The differences between the two parts can be minimal and often overlooked, this inspection also checks other critical elements such as surface finish, and features such as holes.

As part of the workflow, parts are sent out to external contractors for the application of a surface coating and when they are returned they are all together in a shipment that will have a range of parts mixed together. At this point an operator removes the parts for the shipment and sorts them into their specific part type/number before it is sent to the final customer.  Parts that are incorrectly shipped to the customer will lead to substantial fines for the Takumi, so every effort is made to ensure this does not occur. However, this inspection process adds time and cost to the workflow and in order to remain competitive it is advisable to streamline the process and make it more robust. 

Automating the process could assist, however it comes with numerous challenges, namely due to the size variance of the parts (30mm – 2000mm) and the multiplicity of the part types to be inspected (usually in the hundreds)

At the end of the machining, coating and possibly assembly, there is a need for a quality check which is an overview of the completed part. This check generally takes into consideration things such as surface conditions, engraving being present, all features present and identify major defects. There is also a need to sort parts correctly which can be tedious, very manual and can take a lot of time and prone to errors – part mix up. A typical example would be parts returned from a coating process which could have a wide range part parts mixed. These need to be sorted and as this is being done, visual checks are carried out on the quality of the coating process. The process is not high volume and there can be a lot of variation within the components

State of the art

There are multiple solutions available on the market that use 2D images to create and find shape models for object identification. It can be improved further by adding feature inspection like hole diameters and edge distance measurements. Three of the main commercial alternatives are HALCON (https://www.mvtec.com/products/halcon), Adaptive Vision Studio (https://www.adaptive-vision.com/en/ ) and DataLogic System https://www.industrialcontrol.com/datalogic-t-series

A typical development environment for part classification looks like the below, in this case showing the DataLogic solution:

Object identification

Samples

The following pictures show typical examples of aerospace parts, provided by Takumi engineering  (takumiprecision.com). Several sample parts were provided by Takumi and represent a typical sample of parts which need to be addressed under this type of solution.

Model 1 left and right:

Model 2 left and right:

Model 3 Left and right:

Configuration:

IMR’s existing vision system which consists of Halcon software, and an industrial camera were set up in a frame with integrated lighting from all four sides to reduce shadowing. The software was run on a standard laptop.

Process Flow

Below is a typical workflow for the Halcon system. The offline phase is used to introduce new parts into the system.

Results

Over the course of the experiment, the sample parts were introduced to the camera area, and each was successfully recognised. Each part was placed in various orientations without issue. It was noted that for each part, where the part could be placed on different sides, multiple images might be needed but this again is possible. 

Summary

It is a challenge to develop a system that can inspect multiple types of parts of varying sizes with multiple different features combined with a system that is simple to teach when a new part is introduced. There are many unknowns at present, both in terms of technology and software, and the application itself that needs to be identified. This technology sheet has identified the following:

• A system can be built to identify the multiple range of parts.
• The system can be easily adapted to inspect new parts.
• A system can be built which is cost effective.