APPLICATION FOCUS: PROCESS MONITORING g
highly valuable adjunct to the various optical methods just described, because it provides a view of what is happening beneath the surface (where light cannot penetrate). This is especially of interest in keyhole welding, in which much of the process occurs beneath the part surface. Acoustic sensing has
proven a sensitive and reliable real-time indicator of several weld parameters, particularly penetration depth, and void and spatter creation. Acoustic sensing is usually accomplished using either microphones on the processing head, or piezoelectric crystals contacted on the part surface itself. The piezos directly transduce mechanical vibrations into an electrical signal.
It’s a learning process For the most part, these process measurements are relative. The main exceptions to this are monitoring of spectral lines which have fixed wavelength, or pyrometer measurements that can provide an absolute temperature reading. But, usually the acquired data, especially back reflected light and acoustic data, must be compared to a ‘fingerprint’ that has been previously established for that particular process.
Building up this fingerprint
requires running a number of experiments to determine the process window and parameters that deliver a result that is within established tolerances. Usually 10 test runs constitutes a minimum for establishing the necessary baseline data. In the past, the process
window and fingerprint were determined by examining the highest and lowest signal curves for various sensor measurements from several ‘good’ process runs. This data was then used
to establish tolerance bands for the various measured parameters. These upper and lower limits could then be used to signal when violations occurred during actual
38 LASER SYSTEMS EUROPE AUTUMN 2021
processes in a go/no go fashion. The availability of
increasingly fast and powerful microprocessors, together with efficient machine learning software algorithms, has enabled a more sophisticated approach for creating process fingerprints. Instead of simply monitoring each process sensor and flagging any of these which go outside their established limits, this method can simultaneously compare the output from various sensors. For example, a machine learning algorithm can simultaneously compare the output from optical and acoustic sensors and use this to develop a process fingerprint based on probabilities, rather than just absolute limits. The goal of this is to identify problems that occur even when no one specific parameter is outside of its limits, and, conversely, to avoid scrapping or reworking parts in which a limit was transgressed during production, but which are in
“Laser process head manufacturers are using
Readings from multiple sensors can be combined to enable better control of high-precision welding
cost effective, has also been occurring over the past few years. But, the most advanced commercial process monitoring systems have all targeted high-power laser welding using continuous wave (CW) lasers. Now, technology and
machine learning algorithms to combine and analyse data from multiple high- speed sensors”
fact ‘good.’ Plus, these systems can learn continually, using the results of every process run to add to their knowledge. This further ensures that process quality improves and costs drop.
To be precise All the various sensing tools which have been described have been around in one form or another for many years. Plus, the application of machine learning to create process monitoring systems which deliver better results, and are more adaptable and
products are being introduced which specifically service high-precision tasks, such as microwelding, microcutting and marking. Specialised process sensing systems are required for these applications, because the sensors used for high power applications can’t readily be adapted to the requirements of these precision tasks. And, importantly, the sensors, processing electronics and analysis software must allow sampling at a high enough speed to work properly, with pulsed lasers operating at high repetition rates. Figures 1 and 2 demonstrate
how a pulsed laser system, designed and constructed at Coherent within an internal research project for just this purpose, can accurately pinpoint weld defects on- the-fly. Specifically, the slight deviation in the weld bead, circled in figure 1, is identified in both the plasma and temperature signals. In figure 2, a laser beam was scanned perpendicular to the
direction of the weld seam. The part itself was purposefully tilted so that the laser would eventually go out of focus. In this case, signals from the plasma, laser back reflection and acoustic (microphone) sensors all indicated when this occurred.
Conclusion Online process monitoring saves time and money by increasing yields, improving quality, and reducing scrap and rework. It reduces machine downtime by identifying the need for preventative maintenance early on, while also providing a better level of traceability, which is particularly important in industries such as medical device manufacturing. It’s also a key element in making processes Industry 4.0 compatible. Now, laser process head manufacturers are using machine learning algorithms to combine and analyse the data from multiple high-speed sensors to bring these benefits to the most demanding, high precision laser materials processing tasks. l
Florian Furger is research & development project manager; Markus Danner is a product line manager; and Roland Mayerhofer is a product marketing manager at Coherent
@LASERSYSTEMSMAG |
WWW.LASERSYSTEMSEUROPE.COM
Coherent
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46
