In-Line CT Provides Real-Time Quality Feedback on the Production Line

The moving assembly line has proven its worth for over 100 years, ever since Henry Ford introduced his breakthrough in time and motion technology at his Michigan automobile plant in 1913. The introduction of robots (GM, 1962) in addition to human workers has only increased efficiency.

However, the success of an assembly line depends heavily on the quality of the parts that go into the production line.

Therefore, as the complexity of modern automobiles, especially electric vehicles (EVs), has increased significantly in recent decades, the value of pre-assembly quality checks at critical stages in the ongoing production process has become increasingly apparent. In addition to various surface scanning methods, some major automotive manufacturers (along with others in the aerospace, energy, and electronics industries) are now using computed tomography (CT) to deeply inspect the interiors of complex parts to ensure they are suitable for assembly.

CT scans are very expensive these days, but so are expensive parts that are found to be defective after the car has left the assembly line. Quality testing of valuable auto parts before they are installed in the vehicle can save a lot of time and resources.

That’s the conclusion German luxury car manufacturer BMW has concluded, which is why it is carrying out CT scans of the engine housings of all models in its cutting-edge iX series of electric cars.

Unlike the company’s other current electric vehicles, the iX is based on an independent platform. But it uses the same powertrain (motor, inverter and transmission) called HEAT (Highly Integrated Electric Powertrain) that’s used in other BMW models. Ensuring the consistency and quality of this precious powertrain is clearly a top priority for the automaker.

Robots, Scanners and Software – Online

BMW is using an in-line robot, a combination of CT scanning and detection, analysis and visualization software to inspect HEAT electric motor housings during production at its Landshut plant in Germany. The entire system was installed and integrated into the production line by Heitec, a global engineering solutions provider for multiple industries.

Heitec’s mounting system provides a continuous process in which a robot picks an aluminum shell from the production line, places it on a platform that feeds into the HeiDetect inline CT scanner, then moves to the other side of the scanner to pick up the scanned component and place it back on the line. Each scan takes 50 seconds. Watch a video of this process here:

HeiDetect FX (HDFX) Inline CTplay

Video source: HEITEC

Once the scan is complete, Heitec’s proprietary software rapidly creates a digital model of each case from the CT data and transfers the model to Hexagon’s VGinLINE software for rapid segmentation and analysis. It automatically generates graphic and color alerts that engineers can view on-screen, indicating porosity, cracks, geometric inconsistencies or other manufacturing or material defects that deviate from BMW’s specific measurement requirements and could affect quality.

Promptly provide accurate production information

“It’s difficult to do all this within the cycle time of a production line,” says Heitec co-founder Christian Abt. “To get within that time, we have to speed things up, or even just shave a few seconds off of it.”  

Achieving this kind of detailed industrial automation is now second nature to Heitec, which has been installing increasingly “smart” factory systems around the world for more than 26 years. After earning his engineering degree, Abt first worked at Germany’s Fraunhofer Institute for Robotics, “a great playground for young engineers in robotics.” With robots becoming an everyday commodity, he left to co-found Heitec, determined to enhance the young company’s automation offerings by adding CT scanning.

“Prior to that, most CT was used in laboratories,” he says, “so we saw an opportunity to put these machines into production and support the guys in blue suits on the factory floor with our knowledge of x-ray technology, robotics and automation.”  

But adding CT to a production line also brings its own challenges.

“Manufacturing companies that hire us to perform online CT today have often looked at other options and found that no other solution could give them the level of quality assurance they require,” Abt says . “They want answers that work for them.”  

Heitec installs robots and CT scanners to customize every setup for each specific industrial customer, delivering test results as quickly as it promises.  “If you’ve used industrial CT before, you may remember hours of scanning and sifting through gigabytes of data looking for errors,”Abt says . “Not in manufacturing. You don’t have time for those kinds of delays.”  

The secret is in the scan and the software.

The solution to the lack of time? With lower resolution industrial CT scanners, the time it takes to capture an image is significantly reduced. Furthermore, Heitec’s scanner integration software is designed to process this “light” information and reconstruct the volumetric data of the components into a 3D computer model. This data is then loaded into comprehensive visualization and analysis software, which uses complex algorithms and, more recently, deep learning to interrogate the digital model and reliably identify, interpret and report potential quality issues in a much shorter time frame, even in noisy images.

 “We turned to Hexagon’s VGSTUDIO MAX and VGinLINE for advanced data processing software that could give our customers the measurement results they wanted,”  says Abt. “We’ve evolved from software developed for research purposes to a situation where the industry requires more standardized, pre-validated digital tools. Everyone knows the VG software; it continues to be the standard in CT software.”

“VGinLINE’s complete infrastructure combines accepted standards with flexibility, as it is a ready-to-use software package that allows easy integration of your own modules and functions. Just drop the mass and VG starts the analysis. The process can also be easily parallelized and accelerated using more computers, in cases where the data analysis time is slightly longer than the part scan time.”

VG CT data analysis and visualization software has been developing and evolving for over a quarter century, with process automation and, more recently, deep learning and machine learning increasingly being integrated into products.  “Making the most of the vast amounts of design and engineering data that can be collected with CT requires a comprehensive approach combined with powerful computing capabilities,” said Dr. Daniela Handl, general manager of VG products at Hexagon. “Fast delivery of results is crucial for engineers who want to understand in real time how production line quality or variations in manufacturing parameters affect results.”

The key to determining meaningful data in component scans is segmentation, i.e., the extraction of regions of interest (ROIs) from the 3D image data. Abrupt discontinuities in voxel grey values ​​often indicate edges that define a region. By dividing the scan into separate regions, the software can quickly focus on and catalog the characteristics of the voxels within each segment. Segmentation therefore significantly reduces processing time by allowing only the important parts of the dataset to be processed.

Customize your own solutions with machine learning

But how do you determine which segments are important?

Especially with lower-resolution CT scans taken on production lines, the less clear image slices can be more difficult to interpret. This is where artificial intelligence (AI) and machine learning come in. Using trainable algorithms, AI can be leveraged to process and interpret noisy CT data just as accurately as high-resolution data.

The key word here is “trainable.” The algorithm compares what it sees to a pre-existing database of all identifiable defects, customized for each manufacturer’s product geometries (this is where the “training” comes in). The algorithm “learns” to identify specific manufacturing defects by interpreting what the low-definition voxels indicate.

Over time, deep learning gets better at what it sees visually against real product data that the solution already knows, and it learns to recognise patterns and features and point out deviations from the norm. In this way, it can provide a highly accurate snapshot of what is actually happening on a particular production line, so confident decisions can be made about whether to accept or reject a part. This informs changes in production variables so that their impact can be understood, matched and statistically tested. EV batteries are another area where machine learning can be used in this way.

BMW is now exploring ways to optimize this AI approach, along with other leading manufacturers across industries, who are working closely under NDA with innovators such as Heitec and Hexagon to create and manage their own internal datasets to train their in-house deep learning systems.

 “Training a deep learning system with your own data naturally takes time,”  Handl says, “but the payoff will be significant in terms of saving valuable time and resources on the production line.”Actionable insights into what’s happening on the factory floor will contribute to the further development of smart manufacturing, enabling companies across many industries to identify ways to improve quality while making their products more competitive.