The Basque Government has approved the FATIMA Project, which, within the ELKARTEK call, leads AOTEK, the corporate R&D unit of Fagor Automation, in cooperation with the IDEKO and ETIC technology centers and the MONDRAGON UNIVERSITY.
This project is framed in the use of Artificial Intelligence or supervised learning techniques applied to inhomogeneous large-scale data, in which the core of techniques encompassed in the term “Big Data” are required.
The challenge is framed in level TRL4 for prototype-level validations.
The problem being addressed is optimizing the execution CNC programs with High Speed Cutting (HSC) algorithms in the FAGOR numerical controls in the terms desired by the user. This is a difficult problem with a number of facets, in which the remarkable complexity of the reconstruction algorithms of the original geometry and the management of a large number of blocks to achieve a high machining rate is joined by the huge variability introduced by the kinematic trains of the machines and the parameterization of the management of paths in the CNC.
For that reason, the study has been broken down into two different use cases to address it with any guarantee of success. In the first case, casuistry is approached from the standpoint of theoretical paths and simulation on a machine model that, in a first approximation, is assumed to have a linear behaviour, though not necessarily one-dimensional.
This first case study proposes a representative set of parts. It should be carried out on a finite, but large enough, number of interpolation parameters (HSC), with an evaluation against a particular machine (model) with a few adjustments of certain variable links. A first approach, after the project has already been started, has required several computers on which we installed the CNC8065 FAGOR simulator, operating continuously on a set of over 100 HSC-type programs with different characteristics, with several sets of parameters and simulations on several dynamics (machines). Obtaining the first data sets in these conditions took several weeks and has filled up, so far, over 2 Terabytes.
The analysis by FAGOR and its collaborators (IDEKO, ETIC, MGEP) should determine the program features that are relevant to achieving the best possible machining.
The second case addresses the achievement of the best possible model in terms of HSC machining, which can be included in case A. To do that, some of the above programs will be selected and run for various sets of the previous adjustment parameters in a real machine provided by IDEKO, which will also contribute, along with its experience in machine design, with accelerometers and sensors that characterize the behaviour of the tool tip, which ultimately determines the quality of machining. Obviously, despite being very large, the amount of data in this case can’t not reach that obtained in the previous case, by lack of material time, since the execution is carried out in real time, and the simulator is many times faster than the machine. The actual data files, including acceleration data, are stored for further processing in a similar way as in case 1. The system, as designed, should fully engage both scenarios so that the models obtained in the machine are fed to the algorithms that are obtained in the theoretical simulations. The result, which could be referred to as the best set of parameters for said program and machine, should be checked for optimality when running on it.
The final goal of the project is to obtain an AI algorithm that advises the user of the FAGOR numerical controls in the achievement of an optimal machining that is suited to the user needs for its machine . The project aims precisely to obtain help tools for both environments that facilitate the obtaining of more precise machine models and parameterization for HSC algorithms, as well as providing a set of optimum parameters depending on the machining strategy (shortest time, best surface quality, highest accuracy).
For the manufacturer, given a particular machine, FAGOR would provide the ability to adjust the parameterization depending on the type of programs being run. For each type, a set of drive parameters would be recommended and stored in the CNC.
When the end user chooses the program to run, the AI algorithm should decide which of these sets is best suited for the selected strategy.
These are, as can be seen, very ambitious goals, with a remarkable degree of innovation over the current state, with full integration of Big Data analytics and AI strategies, belonging to a high, but focused, level of research on the needs of FAGOR customers.
