Multiple cameras ensure engine-cover integrity

Machine-vision system inspects engine parts for defects.

By Gerald W. Budd

Quality control and high throughput are of paramount importance for Daimler Chrysler, which produces more than 900,000 engines per year, including the Modular OHC 3.7L V6 and 4.7L V8 engines that share many of the same components. Uni Boring, a company that specializes in precision machining, ssembly, and inspection of powertrain products, is an original equipment manufacturer supplier to DaimlerChrysler. Uni Boring precision machines and assembles intake manifolds, cylinder heads, bedplates, covers, and engine blocks, as well as doing partial and full engine assembly for niche and production markets.

Figure 1. Uni Boring uses Phoenix Imaging AVIS to inspect precision-machined engine-front timing-chain-case covers for the high-performance DaimlerChrysler 3.7L and 4.7L models.

One of the high-volume parts supplied to DaimlerChrysler by Uni Boring is the engine-front timing-chain-case covers for both the 3.7L and 4.7L models. These engines are installed into DaimlerChrysler’s Jeep and Dodge vehicles at seven assembly plants worldwide. To inspect these parts, Uni Boring used the Phoenix Imaging automated visual inspection system (AVIS; see Fig. 1). The system performs multiple operations, including inspection for porosity and presence of clear and/or tapped holes, Robot guidance, and optical character recognition for component traceability.

To perform an accurate inspection, it is necessary to create software that encompasses all of the desired inspection tasks. These include inspecting porosity of the “A” and “D” faces; inspecting uncleaned machined surfaces, which may contain casting variations; detecting the presence of through-holes and tapped holes; and providing traceability of every inspected component by means of optical character recognition consisting of an indented marker serial number.

The application required multiple sensors of various pixel resolutions. When initially developed, the resolution of CCD sensors was, for the time, large using 1300 * 1030 * 10-bit CV-M4 Camera Link cameras from JAI Pulnix. This resolution is small when compared to the sensor sizes that are implemented in similar applications today. However, this technology established the standard for on-line porosity inspection and demonstrated its use in a real-world production environment.

Every successful vision application requires a well-designed illumination system. This is even more critical for surface-inspection applications. Phoenix Imaging manufactures more than 300 different types of illumination systems. It designed a special illumination system that could be implemented on-line and that could provide uniform illumination for viewing the machined surfaces of the front cover.

The system inspects the front cover on a vertical pallet conveyor. By holding the front cover in a vertical orientation, both sides and all features can be viewed at the same time. The sensors are interfaced to the company’s PC-based PVS-100 machine-vision system using two CL1 frame grabbers from Epix. The machined features of the component-that is, the quality of irregular shaped surfaces, holes, and threads-are inspected using specific vision algorithms.

The vision algorithms are created using the company’s VisionMaker software. The software allows each of the inspection algorithms to be developed and tested very quickly and acts as a prefilter to all of the company’s Active-X vision tools. It allows the vision developer to construct the vision algorithm as a list of low-level vision functions arranged to massage the image and prepare it before being analyzed by the Active-X tool.

Unlike many vision products, there is no need to compile the vision algorithm before implementation. The developer (or trained user) can modify a vision algorithm in the production environment without special computers, development tools, or libraries. VisionMaker allows generation and test of vision algorithms in minutes.

The covers move through the inspection system on a pallet conveyor system. The inspection system acquires images of both sides of the front cover simultaneously while it is held on a transfer pallet. The software locates the position of the component within the field of view (FOV) of each sensor. It is important to compensate for both component translation and rotation because the pallet may experience wear and tear during normal use.

After components are located, a number of inspection zones can be applied to specific regions. Each inspection zone can apply its unique inspection criteria. The inspection system can perform tasks that human inspectors find difficult, including accurately measuring the size of imperfections, counting the total number of imperfections, determining the density of imperfections per unit area, and determining the proximity of imperfections with respect to each other.

Using the CV-M4 medium-resolution cameras, it is possible to check for porosity defects of <1 mm with a FOV of about 400 * 400 mm. The large-format or high-resolution cameras allow the machine-vision engineer to use one camera in place of four or more cameras, simplifying both system setup and calculations.

The resolution of a machine-vision system can be determined by the ratio of the number of pixels in the sensor to the FOV. Cameras with a pixel resolution of 1300 × 1030 to 2000 × 1600 are considered medium-resolution formats. Cameras with a pixel resolution of 3000 × 2000 or greater are currently considered to be high-resolution format. The latest-generation system has demonstrated the ability to detect porosity of <400 µm on a 500-mm-long automotive component using an 8-Mpixel camera. There are larger-format cameras in the development pipeline that should be available in mid-2005.

Figure 2. To inspect engine parts, the Phoenix Imaging PVS-100 uses several JAI Pulnix Camera Link cameras in a master/slave configuration. Hosted by PC-based systems, the system also uses Phoenix Imaging VisionMaker software and the company's lighting to perform tasks such as porosity and part-defect inspection.

To perform the inspection of features in other regions of the casting, six 752 * 582 * 8-bit CV-M50 low-resolution cameras from JAI Pulnix are used. The low-resolution cameras are interfaced to a second PVS-100M using three SV5 frame grabbers from Epix. The inspection system is designed using a modular format in which one image processor acts as the cell master and additional image processors serve as slave devices. The image processors communicate with each other using a high-speed Ethernet link at 100 Mbytes/s or 1 Gbyte/s (see Fig. 2).

The AVIS acts as the center of the inspection system where all parameters, tolerance limits, and results are displayed. The operator interface is generated using Microsoft Visual Studio, usually programmed in VisualBasic. The VisualBasic environment is preferred because it allows screens to be rapidly customized to meet customer requirements.

The actual inspection process starts on a Viewport-a screen that displays the image as acquired from an individual sensor. Each sensor requires at least one Viewport. The image-processing tools are provided in an Active-X format. The specific tools required for each task may be added by simply dragging and dropping them on the Viewport. The Active-X toolset includes utilities such as alignment, blob characteristics, and curve fitting, which all permit the developer to select the functions that can be used individually or together to measure the specific features of a component. The Phoenix Imaging team developed a system that could be used to classify specific defects in a number of zones, each with a specific set of requirements.

Features, advantages, benefits

“To ensure the integrity of these engine parts,” says Facundo Bravo, Uni Boring president, “each surface must be carefully examined for all cast and machined deformities and sealing face porosity defects. For this, we called on Phoenix Imaging to develop a custom-built image-processing system. Moving these inspection tasks from human vision to robust >machine vision ensures defects won’t reach our customers, while providing traceability into the vehicle.”

The system, with a cost of between $150,000 and $200,000, has paid for itself in less than a year, he says. “When you consider the cost of on-site sorting after a suspect product enters the production stream, such costs become immediately justifiable.”

Company Info

DaimlerChrysler, Detroit, MI, USA; www.daimlerchrysler.com
Epix, Buffalo Grove, IL, USA; www.epix.com
JAI Pulnix, Sunnyvale, CA, USA; www.jaipulnix.com
Phoenix Imaging, Livonia, MI, USA; www.phoeniximaging.com
Uni Boring, Howell, MI, USA; www.uniboring.com