Experimental Results of a Self Learning Compensation System for High Precision Manufacturing

Silvestri, Marco and Fontanesi, S and Carnevale, M (2019) Experimental Results of a Self Learning Compensation System for High Precision Manufacturing. International Journal of Innovative Technology and Exploring Engineering, 9 (11). pp. 2632-2639. ISSN 2632-2639

[img] Text
IJITEE_silvestri_K19490981119 ok.pdf

Download (768kB)


This paper presents the experimental results obtained by applying the results of some recently concluded European projects, whose objective was the development and validation of hardware and control systems for production, based on understanding, evaluating and controlling the performance of a machine tool. It is based on a self-learning controller capable of managing a large quantity of data acquired by sensor systems, as well as on-board artifacts and finished work piece measurements that, associated to operating conditions, permit the accumulation of knowledge regarding the behaviour of machines. Relying on this experience-based approach, the controller can predict the errors that a machining process will present under different conditions and can thus adapt compensation tables. The approach set out has been implemented in a demonstrator consisting of a 5-axis high-precision boring machine, fully functional in an industrial shop floor but used under controlled environmental conditions (thermostatic chamber and special machine foundations). Its software system supports measurement procedures and is able to integrate data acquisition from different sensor systems, to calculate the volumetric error with 3D representations, to provide models for calculation of error functions and to integrate communication processes with the CNC. It can therefore operate in actual production sites, introducing relevant improvements in the machine tools manufacturing field. This paper presents the experimental results obtained during the project validation, including a comparison between on field measurements and compensation tables calculated on the basis of the predictions of the self-learning system. The analysis of the data gathered highlights the system's capability to deal with both simple linear dependencies (e.g. between error and mass variation) and complex, non-linear but repeatable trends. Results discussion, relating to two different and independent axis, demonstrates the applicability of the system under real operating conditions.

Actions (login required)

View Item View Item