On-Machine Measurements are a key factor for shorter closed quality control loops in industrial manufacturing. Especially for the production of large components, they promote the first-time-right approach, which is highly desirable, due to small quantities and steep value chains. In contrast to measurement rooms for CMMs, the production environment conditions are unregulated and impact multiple factors along the on-machine measurement metrological chain. As presented as a keynote speech at the XXXI CIRP Sponsored Conference on Supervising and Diagnostics of Machining Systems "MANUFACTURING ACTIVE IMPROVEMEN" by Professor Dr. Robert H. Schmitt, this article reviews current research and ideas regarding on-machine measurements. The authors collect necessary process data with the help of new technologies in the course of digitalization and thus propose a holistic model for systematic error compensation and measurement uncertainty prediction. For assessing the machine's volumetric accuracy under thermal loads, the authors develop a novel modelling approach, which determines transient geometric errors by abstracting structural parts as spline curve with typical deformation modes. To address the workpiece's influence on the measurement process, a data-driven framework, fusing realtime sensor-data with the virtual component, is used to model and predict transient thermo-mechanical workpiece states. For dissemination, the authors continue working on ISO standardization and, as subjects of future research, explore new paths in terms of data-driven modelling approaches, using physical abstractions coupled with machine learning and live process data.

Modern CNC machine tools provide lookup tables to enhance the machine tool's precision but the generation of table entries can be a demanding task. In this paper, the coefficients of the 25 cubic polynomial functions used to generate the LUTs entries for a five-axis machine tool are obtained by solving a linear system incorporating a Vandermonde expansion of the nominal control jacobian. The necessary volumetric errors within the working volume are predicted from machine's geometric errors estimated by the indirect error identification method based on the on-machine touch probing measurement of a reconfigurable uncalibrated master ball artefact (RUMBA). The proposed scheme is applied to a small Mitsubishi M730 CNC machine. Two different error models are used for modeling the erroneous machine tool, one estimating mainly inter-axis errors and the other including numerous intra-axis errors. The table-based compensation is validated through additional on-machine measurements. Experimental tests demonstrate a significant reduction in volumetric errors and in the effective machine error parameters. The LUTs reduce most of the dominant machine error parameters. It is concluded that although being effective in correcting some geometric errors, the generated LUTs cannot compensate some axis misalignments such as EB(OX)A and EB(OX)Z. The Root Mean Square of the translational volumetric errors are improved from 87.3, 75.4 and 71.5 µm down to 24.8, 18.8 and 22.1 µm in the X, Y and Z directions, respectively.

The volumetric performance of machine tools is limited by the remaining relative deviation between desired and real tool tip position. Being able to predict this deviation at any given functional point enables methods for compensation or counteraction and hence reduce errors in manufacturing and uncertainties for on-machine measurement tasks. Time-efficient identification and quanitification of different contributions to the resulting deviation play a key role in this strategy. The authors pursue the development of an optical sensor system for step diagonal measurement methods, which can be integrated into the working volume of the machine due to its compact size, enabling fast measurements of the axes’ motion error including roll, pitch and yaw and squareness errors without significantly interrupting the manufacturing process. The use of a frequency-modulating interferometer and photosensitive arrays in combination with a Gaussian laser beam allow for measurements at comparable accuracy, lower cost and smaller dimensions compared to state-of-the-art optical measuring appliances for offline machine tool calibration. For validation of the method a virtual machine setup and raytracing simulation is used which enables the investigation of systematic errors like sensor hardware misalignment.

Abstract:Monitoring of the relative deviation between commanded and actual tool tip position, which limits the volumetric performance of the machine tool, enables the use of contemporary methods of compensation to reduce tolerance mismatch and the uncertainties of on-machine measurements. The development of a primarily optical sensor setup capable of being integrated into the machine structure without limiting its operating range is presented. The use of a frequencymodulating interferometer and photosensitive arrays in combination with a Gaussian laser beam allows for fast and automated online measurements of the axes' motion errors and thermal conditions with comparable accuracy, lower cost, and smaller dimensions as compared to state-of-the-art optical measuring instruments for offline machine tool calibration. The development is tested through simulation of the sensor setup based on raytracing and Monte-Carlo techniques.

AbstractUncontrolled environmental conditions often impact manufacturing processes and lead to product quality fluctuations. For machine tools, thermal influences are a major limitation to the volumetric performance. Climate controls for the shop floor, and machines, or thermally stable structural designs are economically not feasible, promoting control-based compensation as a possible solution. Since the relationship between disturbing quantities and effects are complex and specific to each machine, appropriate modelling is a critical requirement. The authors describe an approach for developing hybrid models, superposing white-box model knowledge, and machine learning. The overall effort can be optimized by combining and balancing different modelling methods, like designing the physical model part and training intelligent algorithms. A general model structure allows a continuous integration of different white-box and black-box model components. The authors integrate self-developed smart sensors into a demonstrator machine tool to test and validate the performance of the approach.