Automated Automotive Body Inspection System for Poka-Yoke Production Lines Using NI IMAQ Vision

Author(s):
Ferencz András - Naturen KTF.

Industry:
Automotive

Products:
Vision Development Module, LabVIEW

The Challenge:
Developing a machine vision inspection system for inspecting finished automotive bodies (“white bodies” or “bodies in white”) prepared for painting at the end of the frame assembly line in the welding shop.

The Solution:
Deploying an inspection system that can decide if all necessary components have been mounted on the body under inspection and, in cases where missing or faulty components are detected, the system can remove the body under inspection from the assembly line so it cannot enter the paint shop.

The line controller located at the start of the assembly line stores the body number type information for each body by reading an RFID stamp. All required body components are mounted on the body as it travels from the start of the assembly line to the inspection location at the line’s end. The body passes between rails at the end of the assembly line where the inspection cell of the machine vision system is installed. The finished bodies are automatically checked for the presence of and compliance to the type specification of each necessary component.

In the first step of the inspection process – before the body enters the inspection cell – the system acquires the body number and type information stored by the line controller. The body number and type information are reference data necessary for selecting the specific inspection sequence that should be performed for the body under test. The body travels the assembly line on a conveyor that hauls it using pull rods.

Once a body enters the inspection cell, the system tracks its position using a rotary encoder coupled to the pull rod drawing the body. Inspection time is limited by the time a body spends in the inspection cell. In this relatively short period, the system takes several images of the body from different angles. Using these images, the system checks for the presence of all necessary components and takes more images of the recognized components. Then the system produces a log of the measurement results and moves all the images to a high-capacity redundant storage.

Finally, the system decides on if it is necessary to remove the body under test from the line. The cycle time allowed for carrying out all tasks is about 35 seconds. Because of the short and fluctuating cycle time, an important consideration is task scheduling. Different inspection tasks should be queued and timed in such a way that they can be completed before the next body arrives. We have addressed this challenge through asynchronous image acquisition, evaluation, and storage. Depending on the inspection result, after leaving the inspection cell, the body is either removed from the assembly line by a lifting mechanism or is passed on to a sorter that prepares it for the paint shop.

Apart from the short inspection time, several other factors had to be taken into account during system design. First, the bodies move along the assembly line with a varying but limited translation speed. Also, frames may undergo small-scale tilting or rocking movement in directions perpendicular to the direction of travel, and the assembly line may be stopped and restarted at any time. Component recognition was rendered difficult by tilting and rocking movement. The uneven surface quality of sheet metal (scratches, inscriptions, and so on) also put obstacles in the path leading to the desired 99.9 percent recognition rate.

Operator access to the body under inspection had to be maintained, which meant that we could not implement the inspection cell as an enclosure blocking all ambient light. To work around ambient light problems, we used infrared lighting. Infrared lighting is superior to using bright white light also from ergonomic aspects because we could avoid operator discomfort caused by harsh white light. The infrared light sources are switched off not only at the end of each inspection cycle but during shorter pauses between steps of the inspection process, and even when the assembly line is stopped during an ongoing inspection.

The system displays inspection results per component type on a PC screen. An image viewer feature is provided for viewing images of individual components for double-checking or result verification purposes. The system receives and sends data to and from the assembly line controller, values of communication variables are used for providing communication between control units, programmable logic controllers (PLCs) and compact vision systems (CVSs), and memory and disk space status data can also be displayed. Display and off-line data evaluation parameters may be adjusted as well. According to a design requirement for the system, automated inspection should be able to be operated independent of the PC display.

Integration of NI Components into the Inspection System

The presence and/or absence of specific components located on the left and right sides of the frame as well as on the roof cover plate should be checked irrespective of car type. Because the inspection can be performed at the same time on the left and right sides of the body, two CVSs (one for each side) are applied for image evaluation and component recognition. The inspection of components located on the left and right sides of the roof cover plate and on the front and back of the car frame are allocated to the CVS on the corresponding side.

We had to use two separate CVS units because a single unit could not have handled all seven inspection cameras installed in the system. The CVS hardware provided the computing capacity necessary for performing the machine vision inspection tasks, while the NI LabVIEW platform was used for developing the software.

Special Benefits of NI Components for the Inspection System

The above mentioned NI components all have the potential for enabling the development of industrial machine vision and image processing algorithms. We implemented the principal function of our system, the evaluation of images taken of automotive frame bodies, using those components. Using the NI Vision Assistant and NI Vision Builder for Automated Inspection, we could easily and quickly develop solutions for recognizing different frame components. We could use several different images to verify the efficiency of the algorithms, and we could always check intermediate results during the development process.

Recognition solutions developed with the help of the NI Vision Development Module were implemented in LabVIEW (either by means of automatic code generation or writing code manually) and then integrated into the system running on the CVS. The NI CVS-1450 Series offered a compact solution for image processing, communications, memory management, and other computation-intensive tasks. With NI CVS-1450, we could carry out synchronized image acquisition and position measurement using rotary motion encoders. Thanks to the combined application of these NI components, we created a system that operates reliably in spite of bottlenecks caused by the short cycle time and constraints on available memory.

Machine Vision Inspection for Reduced Production Costs

Using our machine vision inspection system, we can recognize and remove from the line automotive bodies having components not conforming to type specification mounted, or those lacking one or more component. The end-of-line inspection helps identify those bodies that have to be sent back for repair due to faulty or incomplete assembly.

When the operators perform the visual inspection manually, the “when and where” of detecting an assembly fault cannot be determined in advance. For instance, it may happen that a body has already been transferred to the paint shop when an assembly fault is detected. In this case, a costly return to the welding shop is necessary. Still worse, assembly faults may go undetected until painting is completed or even until the customer receives the completed car. This may necessitate transporting the car back to the factory and repairing and repainting it – these operations have significant additional costs and take up a lot of valuable production time.

Prospects for Further Development

The system’s software architecture was designed to allow the inclusion of new inspection tasks needed for testing new car types. With the flexible hardware architecture, we can connect more CVSs or cameras to the system in the future. Thanks to the carefully thought out memory management and CPU scheduling, new memory-intensive image processing tasks may easily be implemented and integrated into the present system. Further expansion of the system by implementing new component-recognition tasks is also supported by asynchronous image acquisition and asynchronous evaluation and logging.

Author Information:
For more information on this Case Study, contact:
Ferencz András
Naturen KTF.

Browse All Case Studies »