Industrial Machine Vision Based on National Instruments Technology Goes Further in Complex Human Task Automation

Author(s):
Pascal Cormier - Avera/Vision & Robotic
Sebastien Parent - Averna/Vision & Robotic

Industry:
Industrial Controls/ Devices/ Systems, Medical/ Medical Instrumentation, Manufacturing, Life Science

Products:
PXI/CompactPCI, LabVIEW

The Challenge:
Reproducing complex human inspection tasks with multiple visual inspections with a high level of precision (+/- 0.001 inch) and a high level of complexity (defect detection on highly reflective metallic surfaces with complex shapes and validation of pin stamped codes) for various part models and production by small batches.

The Solution:
Using a robot to handle and move the parts, multiple images are acquired on three acquisition stations. The system is based on the National Instruments LabVIEW 8.20 platform using NI Vision Acquisition software for processing speed optimization. The robot gives the flexibility needed for various part models and the adaptability for frequent production change overs.

In some manufacturing processes, such as medical and aeronautical, parts are made with high precision and require critical surface quality. These inspections are usually left to human inspection because of their complexity: highly reflective metallic surfaces to be seen at various angles and various lighting conditions and part shapes not entirely visible with a static camera and lighting assembly. Averna Vision & Robotics developed a system that mimics the human inspection strategy for these applications using a robot, a combination of lighting setups, and multiple cameras.

Automation is usually involved when a specific product is to be produced in a large volume and for large batches, but there is also a need to reduce the human subjectivity in complex visual inspection tasks for small production quantity and frequent change over. The level of complexity in automation is highly relevant to the level of flexibility needed for the automated system. Specifically, in machine vision inspection, complex product shapes with high reflective finishes and critical surface quality, as artificial human replacement parts or aircraft motor engine components, are good examples of these applications. The difficulty comes from the efficiency limit of using static setup of cameras and lighting to see all the characteristics of the part to inspect. 

In this paper, we present a system developed by Averna Vision & Robotics to perform such complex inspection on metallic parts. To simulate the operator moving the part in different orientation for light effect variation, we use a robot with a gripper, a combination of cameras and lighting. The design is done in a way to give flexibility for a large part range (dimension and shape). The system is also designed to process small batches of products and accommodate the learning of new products in the future. The National Instruments platform was selected for the software and the hardware.

System Design

The system architecture is PC-based with two NI-PCIe-1430 Camera Link image acquisition boards with dual port. The inspection sequence and the user interfaces (viewer, alarm, and calibration) were developed in National Instruments LabVIEW 8.20. For every image processing task, the NI Vision Acquisition library algorithms are utilized. The communication with the robot controller FANUC RJ 3iB is performed via an NI-PCI-6514 I/O board.

NI LabVIEW was selected for its multitasking capabilities that simplify the management of parallel tasks and also for its capability to easily integrate control and machine vision in one development platform.

The major steps of the inspection system are:

• Learn or select a product
• Localize the part with the robot
• Move the part to perform the various inspection tasks

One of the critical steps is the precise localization of the part in the FANUC LR Mate 200iB robot world. First, the part is roughly inserted in a part holder. Once in place, the robot picks the part with high precision using the gripper. Once the part is in its hand, the robot goes through three different image acquisition stages: the optical character validation station, the random surface defect detection, and the dimensional measurements.

Optical Character Validation (OCV)

The purpose of this inspection is to validate that the part model and the batch number marked on the part is the good one. The actual system is designed to evaluate characters made by pin stamping directly on the part surface. This kind of marking is difficult for an Optical Character Recognition (OCR)/OCV algorithm because each character is made of dots instead of a continuous line, but the NI IMAQ OCR VIs worked well for these types of characters.

The components of the acquisition system are an LED ring light from Boreal Vision and a high-resolution camera JAI CV-M4CL from JAI PULNIX. The camera is connected in the first Camera Link port of a NI PCIe-1430 acquisition board. Based on the actual camera field of view, the resolution for this inspection is 0.001 inches per pixel.

Random Surface Defect Detection

A large advantage of using a robot in this solution is to have the flexibility to move the parts following six degrees of freedom, giving the capability to see all portion of the part to inspect even if it has a complex shape. The robot is programmed to position the part in front of the camera. Multiple images are acquired with the part tilted and rotated in a way to cover its entire surface at various angles. The principle of this inspection is to learn the state of a good part surface. During inspection, anything appearing different than this learned surface will be evaluated more closely to determine if it is a defect and in which category it falls. For the classification of the defect, the criteria are mainly based on the morphological analysis from the NI Particule Analysis VI. The kinds of defects actually detected are dents, scratches, die marks, grinding marks, pits, and discoloration.

To perform this inspection, a second JAI CV-M4CL camera is used. The lighting is an LED diffuse surface. The camera is connected in the second Camera Link port of the same NI PCIe-1430 acquisition board used for the OCV inspection. Based on the actual camera field of view, the resolution of this acquisition station is 0.001 inches per pixel.

Dimensional Measurement

The last inspection station gives a measurement precision of +/- 0.001 inches. This performance was evaluated with statistical tools such as Gage R&R and linearity analysis compared to values from a coordinate measuring machine (CMM). To reach such precision, the part is scanned through two high-resolution line scans at 90 degrees apart. Using these two orthogonal points of view, it is possible to compensate the robot’s imprecision during part scanning.

This acquisition device is using two DALSA P2-22-04K30 line scan cameras with LED backlights. The two cameras are connected in the ports of the second NI PCIe-1430 acquisition board. The line acquisitions are triggered to obtain the same resolution as the field of view, which is 0.000 5 inches per pixel.

 

This inspection system is pushing the capability of automated inspection systems further in the domain of complex human inspection tasks. The utilization of a robot gives enough flexibility to the system to process a lot of different part models. Because the change over between products is mainly software, it is easy to perform small product batches.

The choice of LabVIEW as a development platform gives the software integration capabilities and optimizes multitasking to keep the processing time very short. At this moment, the targeted cycle time was always under 12 seconds, but it is directly related to the amount of robot manipulation to perform on each part.

For more information, contact:

Charles Magnan

Averna Vision & Robotics

269 Prince Street

Montreal, Quebec

Canada

Tel: 514-788-1420

Fax: 514-866-5830

info@averna-vr.com

www.averna-vr.com

Browse All Case Studies »