• Font Size:
  • A
  • A
  • A

Case Studies

Multi-ROI Use Case: Part Kitting

Neurala, Inc.


As consumers, our expectation for customization has grown significantly. A great example of this is in the automotive industry: we expect an ability to pick and choose the features we want. In order to do this on a modern assembly line, parts need to be put into a “kit” that can be used as a car moves down a manufacturing line. While this can be done with any complex assembly process, where each assembly may have different parts, it is becoming more common within automotive manufacturing. Kitting allows for customization, or can be used when many small parts are used in a single assembly. At the same time, it reduces the complexity on the line and improves efficiency of material handling. It also reduces the chances of the wrong part being picked and used line-side, ensuring that the operator has the materials needed to complete the assembly.

Current State

The kitting process is seeing more robotic automation, but the most common practice still has operators creating the kits. The form factor of the kits can vary based on the type of parts, but they are usually boxes or racks. The kits are often fixtured, so that each type of part is in the same spot in each kit. There is minimal, if any, inspection being performed. Processes are often created to minimize errors (such as picking sequences, using picking lights, and computerized systems), but there is often no inspection to ensure the correct parts are in the kit after the kit is created; it is assumed that parts are being chosen correctly by the kitting operator and placed in the appropriate bin. This can lead to incorrect parts being used, or the operator not having the right parts on the assembly line which can affect productivity.

Neurala VIA Basic Implementation

A GigE camera is added to the kitting workstation. Each potential “kit” has its own anomaly model. A fixture is used to make sure parts are in reliable spaces. The operator tells the system which kit they are pulling for by scanning a barcode on the kitting container, and the barcode indicates which kit is being build and what parts are to be expected.

When parts are not in the kit, or incorrect parts are placed in the kit, an error message is sent to an HMI via Modbus TCP through a PLC to tell the operator to check their work.

In this instance, a model needs to be created for each possible kit variant, which while completely possible, does require a lot of data.

Neurala VIA Advanced Implementation

Building on the basic implementation, we can use multiple regions of interest (ROIs) to simplify the model building and be able to provide better quality metrics. Each area of the kitting fixture can be its own region of interest, meaning that each part has it’s own ROI.

This simplifies model building: each Region of Interest needs to be trained on each possible part variant. But then kits can be made up by selecting the relevant parts in each model, rather than needing to build a model for each kit (reducing the amount of training required).

In addition, rather than just having an indication that the kit is incorrect, using a multi-ROI inspection, the specific part that is incorrect can be identified.

Adapting to market changes

Because demand is down for some consumer goods, manufacturers are slowing down production and decreasing the number of operators on the floor. Many view now as a good time to begin prototyping in order to better solve this problem when they ramp back up again.

The Bottom Line

  • Better inspection, ensuring cars receive the correct parts the first time, minimizing rework. 
  • Ability to catch errors early means less risk of wasted time and resources. 
  • No additional time added to the kitting process.
  • Easy to change/add additional kits. 
  • No need to take the time to build expensive models.


© 2020 Neurala, Inc Boston, MA All Rights Reserved

Search AIA:

Browse by Products:

Browse by Company Type: