Steinbeis experts work with the University of Stuttgart to develop a new kind of algorithm for monitoring milling processes
The competitiveness of manufacturing facilities and companies in high-salary countries like Germany depends on the cost-effectiveness and performance of production machines. In Germany, production sites in the field of machining can only remain viable in the long term if they can automate as many procedures as possible. To do this, robust and reliable process-monitoring systems are required in order to quickly assess the current status of processes and if necessary make corrections. In July 2018, a collaborative ZIM project looking into “the development of a system for monitoring and diagnosing milling processes by analyzing acoustic fingerprints online” came to a successful conclusion. For the project, STASA (Steinbeis Angewandte Systemanalyse GmbH) and the University of Stuttgart Institute for Machine Tools (IfW) codeveloped self-learning algorithms capable of identifying malfunctions and wear during milling processes. The algorithms help record acoustic signals through special microphones mounted in workshops. The project was funded by the Federal Ministry for Education and Research.
Heightening competition is forcing a large number of businesses to reduce the number of people working on their technical systems, which are frequently now highly automated. Many modern processing stations are now completely behind barriers, making it extremely difficult for humans to monitor individual production processes. As a result, machine operators are often not even in the same location as certain processes. Jumping in with manual overrides when something goes wrong is practically impossible, especially given the high traveling speeds and axial accelerations of many machines. With good maintenance and early fault-finding, many machine faults can be avoided nowadays, but unforeseen situations and malfunctions – often caused by the actual machining process – will always be difficult to anticipate. Such anomalies frequently result in quality issues with finished workpieces, or faulty tools and machine parts.
With milling processes, vibrations can occur when the cutting tip of a tool makes its way into a workpiece. These can be measured by using relatively expensive sensors which pick up noises inside structures (“structure-borne sound”), or they can be detected through workpiece vibrations when sounds are emitted into the room where the milling machine is being used. This allows malfunctions to be detected by the human ear. Experienced operators who have worked for some time on the same machine often notice changes in the sound caused by processing, so they know if a process is working as intended or a change occurs in the running noise of a spindle or feed axis.
When a new tool causes wear to a cutting insert, this initially happens quite quickly before leveling out, after which point wear remains relatively stable before peaking again toward the end of the usage cycle of a material. The typical progression of wear and how long this takes depend on the material the workpiece is made from. The algorithm developed for this project makes it possible to determine the status of tool wear by simply listening to airborne sound emissions, which can be recorded using a microphone. In practice, recognizing wear using airborne sound analysis can be completely automated. This makes it possible to react quickly to signs of tool wear by changing cutting inserts or adjusting milling speeds.
Airborne noise analysis also makes it possible to detect cutting defects. The analysis is carried out at specified moments in time so that cutting defects can be picked up in real time. A machine can then be halted to protect undamaged cutting edges from exposure to excessive loads. This does not require a learning and calibration phase before production, so the new algorithm is particularly well suited to production volumes as low as a batch size of 1.
The team working on the project also confirmed that detecting internal shrinkage on workpieces – i.e. undesirable material defects – is also possible using microphones and airborne sound analysis. The algorithm compares acoustic signals of workpieces during processing with reference recordings of sounds previously emitted by workpieces without faults. The skill is to almost completely filter out disturbances caused by acoustic influences in the surrounding area – for example noises from a neighboring machine. Otherwise it can be impossible to detect minute shifts in the frequencies detected by microphone recordings, which are typically caused by material shrinkage. To do this, different filtering and analysis processes are combined, such as software-based lock-in techniques and other noise-cancellation algorithms.
To spot the differences between shrinkage and an intended change to a workpiece, such as a drill hole, reference sounds are required, ideally from the acoustic signals emitted by a perfect processing routine on a milling machine. This allows the algorithm to learn the characteristic features of a perfect process – the timings and frequency ranges of airborne sounds – and these provide a reference point for detecting imperfect processes. Users can set threshold levels for the point at which a workpiece counts as a reject. Compared to simply sorting workpieces manually – shrinkage or no shrinkage – live testing showed that this approach is much better suited to practical application, since shrinkage comes in different shapes, sizes, and extremes. Also, depending on the application, there are times when shrinkage may be acceptable.
The project team at STASA and the University of Stuttgart have introduced the algorithm developed for the ZIM project to hardware and software prototypes for testing in actual machining processes. The experts are currently looking for hardware partners to translate the developed process monitoring solution into a product that’s ready to go into serial production and enter industrial application.
As producing small batches becomes increasingly important and more and more processes go down to single-batch production, it is becoming crucial when using process monitoring systems to avoid long set-up and learning phases. Process monitoring needs to get underway quickly, and the prototype that has now been developed offers a decisive advantage over conventional systems. The new algorithms can start monitoring tools without running a single reference cycle. To monitor workpieces, only one reference cycle is needed. Setting up microphones is a relatively inexpensive and versatile alternative to conventional structure-borne sound detectors, which are also unsuitable for quality monitoring purposes. To filter out sounds coming from the surrounding area, software algorithms can be used, many of which would have been unimaginable even only a few years ago due to limited computational capabilities.