July 19, 2018
Inline monitoring for laser processes
Where lasers weld and solder, inline quality monitoring isn’t far behind. Optical systems track the bonding processes in real time to prevent production faults.
The human eye is unable to keep up with the speed at which laser welding systems work. Scanner heads speed across work pieces making accurate metal seams in a fraction of a second. So that no faults slip in, bonding processes such as this are monitored meticulously in real time. The vision of Photonics 4.0 is already largely implemented in laser bonding processes.
Like the laser processes themselves, inline monitoring is based on photonics. A new approach in this field is optical coherence tomography (OCT), with which, among others, the two affiliated companies SCANLAB and Blackbird Robotics implement 360 degree monitoring. With this method, the topography before, on, and behind the welding point is measured with a high degree of spatial and timely resolution. To do this, the measuring light of the OCT sensor is coaxially coupled directly into the optical path of the laser welding head and positioned via its scan mirror. This enables the entire process area to be monitored—and, as opposed to monitoring from the side, there are no obstacles.
Precise topography data as a pioneer for 4.0 processes
The system checks the edges and the overlap directly before the bonding process and during the process measures the penetration depth of the vapor capillaries and every deviation from the target—and afterwards it checks the quality and position of the welding seam. Pores and craters in the molten metal are detected immediately. To further develop the technology and implement complex processes with oscillating laser beams, Blackbird, with partners such as BMW, Precitec, and the Technical University of Munich are working on the RoKtoLas project within the government funding initiative Photonics for flexible, networked production. The aim is to make OCT suitable for remote laser beam welding in highly flexible car body manufacturing process chains. Background: The automotive industry urgently needs quality assured solutions for the small series production of electric vehicles.
The Scansonic Group, based in Berlin, Germany, also specializes in solutions for remote laser welding and recently announced the fourth generation of its fully integrated ALO SCeye series. SCeye integrates a high-speed CMOS camera system, a switchable laser /LED lighting module, and a computing unit with storage capacity for eight hours of process recording in the welding head. Here, too, the bonding process is monitored and documented from a bird’s eye perspective. With synchronized recording of the camera stream, the process data, part and welding seam numbers, it’s possible to analyze the specific causes of faults and declines in process stability. But there’s more: If the parameters of an optimum process are stored, corresponding machine learning algorithms can compare running processes and recognize faults immediately. Scansonic has evaluated a control system such as this. Even in the validation phase involving 20,000 welded seams with dozens of faults, every single fault was detected. This creates the basis for automated, self-controlling 4.0 processes. Some fine tuning is still required to rule out false alarms and extend the checks to cover defects in the sub-millimeter range.
Interaction of different optical sensors
One approach to increase the reliability of learning control systems is to merge several optical systems. When high-performance cameras and diode systems track the laser processes in the visible and infrared ranges, OCT sensors measure the topography, and when this data is compared with defined process parameters, the path to closed control loops is not far away.
Plasmo Industrietechnik GmbH is working together with Volkswagen, among other companies, to make data from sensor fusions useful for optimizing processes with the help of deep learning software. In this context, it’s important to visualize the information that is relevant for users so it can be used immediately to control production. If the process strays from the set framework, the quality inspection system makes this visible in real time. To do this, big data algorithms in large production networks analyze millions of datasets—and filter the relevant information from these.
Plasmo already has multi-sensor systems such as this in industrial applications. Up to 30,000 images per second are produced where different systems deliver data and parallel measurements of the light emissions, the laser power in the raw beam, laser triangulations, and OCT measurements take place. The data that is acquired makes the effort worthwhile: It not only documents that fault-free welded components are delivered, but on the basis of the data it’s possible to detect and rectify the causes of deviations quickly despite the extremely high level of complexity. In the medium term, it’s likely that inline systems will replace destructive testing. In the long term, they will pave the way for fully automated photonics 4.0 processes.