• Troubleshooting and tuning of vibrating machining processes need dynamic characterization of the machine, which requires the interaction of an expert. This is far from the self-reliant concept.
• Modelling techniques for dynamics characterization are far from the direct use of sensory data structure, although the industry prefers data-driven parameter-less solutions for intervention.
• There is an unnecessary double interpretation of the machine dynamics during the troubleshooting: measuring – interpretation of measurements to humans – human experience, knowledge, modelling and prediction – interpretation to the machine – intervention in the process.
• In most of the cases, the expert’s dynamic characterization is performed in steady machines, although the machine behaves differently during operation.
• The workpiece may have slender/weak parts whose dynamic behaviour significantly changes during machining. This needs to be taken into account for a self-reliant solution.
• Machine tools and industrial robots have varying stiffness along the workspace and often show nonlinear behaviour which significantly changes process behaviour.
• Industry lacks measurement techniques to identify nonlinear dynamics with impulse, harmonic and stochastic excitations.
The project disrupts the ‘double interpretation’ concept of the conventional dynamics characterization techniques used by experts in vibrations diagnostics of machine tools. It aims to develop a framework where the determination of model-related vibratory parameters has secondary or no importance. The proposed framework utilizes a direct and model-less description of the machine tool dynamics from the measured sensory data. As opposed to the parameter-driven differential formalism of cutting processes, the proposed Lendület (Momentum) II project uses the kernel-based convolution formalism of dynamic systems.
- linear kernel description of homogeneous dynamics (IDS Modal)
- modelling of the process that excites the identified dynamics (IDS Process)
- generation of the nonlinear description (IDS Nonlinear)
The proposed concept relies on fundamental and experimental research. The neural network in the IDS Modal module is trained by state-of-the-art multi-core hardware that is able to achieve optimized solution of the supervised learning problem on the collected big data. The entire project is built around the direct processing of sensory data to realize effective actuation actions. On the one hand, the concept in the IDS Process and IDS Nonlinear go further than black-box methods. It develops techniques over the IDS formalism by avoiding (comprehended) model development to predict the dynamic behaviour of the combined real-time controlled machining process. On the other hand, test benches are planned with significant sensory and actuation solutions to validate predictions by mimicking controlled guide, robotic manufacturing and creating simulated nonlinearities in test environment.