Production methods of tomorrow: More efficient tool use through artificial intelligence – ETMM Online

Posted: March 17, 2022 at 2:18 am

16.03.2022A guest post by Mathias Schmidt*

When machining a component, tool wear and the metal removal rate are the decisive factors. Machine learning can provide a valuable contribution to the optimisation of production costs by supporting the decision-making process for tool changes.

Related Companies

(Source: Nikola Krieger)

As with all industrial fields, there is ever growing cost pressure in the machining sector. The more efficiently tools are used, the lower the costs become. However, there are no patent solutions here, individual processes are too different from one application to the next. Transfer learning can offer a solution: Here, knowledge from related tasks that have already been learned is used to train machine learning (ML) models more quickly for new, but related tasks. A research project funded by the German Federal Ministry of Education and Research (BMBF) has been running since June 2021 to explore the possibilities of transfer learning in machining and make it usable for industry.

The production costs of a machined component are largely determined by the metal removal rate and tool wear. With constantly increasing cost pressure, optimising tool use is therefore a promising starting point for reducing costs and increasing efficiency. If tools are replaced too late, the wear has a negative effect on the workpiece quality. In addition to deviations from the required geometric tolerances, increased burr formation, increased roughness and the impairment of the metallurgical and mechanical properties of the workpiece edge zone are consequences of worn tools. Therefore, in industrial practice, tools are often replaced far too early as a precaution. But this also has a negative effect on production costs. In addition to the wasted tool life potential, set-up times and tool costs also increase. AI-supported, intelligent tool management can help to optimise tool life.

By first learning suitable models, it is possible to predict tool wear during machining by in-situ measurement of vibrations, acoustic signals or process forces. Conversely, the expected process forces and temperatures can be estimated with a known initial state of wear. In addition, it is possible to predict the production costs and component properties such as roughness, burr height and the microstructure or microhardness present in the microstructure with a known selection of the process parameters for different manufacturing processes. This means that tools can be used for much longer without the risk of problematic wear. In this way, a resource-efficient and sustainable improvement in productivity can be realised, which can contribute significantly to an increase in the competitiveness of manufacturing companies.

Not all machining is the same, however. In addition to a variety of materials that can be machined, it is always important to consider the process itself. Even with standard tools, there are significant differences. The tools not only consist of different materials suitable for the respective application, but usually have different geometries and possibly even coatings. The results of one application therefore cannot be easily transferred to other applications. Furthermore, training the systems is often very time-consuming. Up to now, available solutions for optimisation by means of ML usually refer to a specific cutting process on a material with defined tools and a defined range of cutting parameters, usually under laboratory conditions. As a result, it is not possible to transfer the models to real, varying machining processes in manufacturing companies using current methods.

Transfer learning can offer a possible solution. Here, knowledge from related applications that have already been learned is used to train ML models more quickly for new tasks or use cases. However, there are still no procedural models that enable transfer learning to be used for applications in everyday industry. This is where the research project "Control of machining processes through transferable artificial intelligence - basis for process improvements and new business models (TransKI)" in the funding measure "Learning production technology - use of artificial intelligence (AI) in production (ProLern)" comes in, which is funded by the German Federal Ministry of Education and Research (BMBF).

The overall objective of the project, the development of transfer learning for the creation of ML models that can be transferred to new fields of application with little effort, was divided into three sub-objectives. The first sub-goal is the determination and modelling of causal interactions in machining. The second sub-goal was defined as ensuring transferability, which finally results in the third sub-goal of making the models usable.

When choosing ML models for machining, it is always important to consider the process itself.

(Source: Nicola Krieger)

In the first phase of the research project, industrial use cases are defined, and machining tests are carried out and evaluated. With the processed data from these tests, basic ML models can be developed. The second phase is about making the models suitable for new use cases. In this process, the test environment, i.e. the process, the machine and sensor technology as well as the material are changed step by step, wear-dependent commonalities are identified and expert knowledge is included in the investigations. In the third project phase, an assistance system for process pre-control and transfer-learning-based business models will be developed to make the optimised ML models industrially usable.

The knowledge gained will be validated in several heterogeneous pilot applications for drilling and milling. Furthermore, the project not only addresses the specific problem of the tool industry, but also opens up new ways through transfer learning to leverage previously untapped value creation potential, for example with capital goods manufacturers and manufacturing companies in other sectors.

This forward-looking and comprehensive project requires expertise and resources from various fields. That is why a total of seven partners are involved in the joint project. The experts for precision tools from K.-H. Mller Przisionswerkzeuge are coordinating the project and are responsible for the development of innovative, AI-based business models. Robert Bosch is examining the transferability of the ML models to industrially relevant milling processes and is contributing existing experience in the use of AI/ML methods in production technology to the project. As an industrial partner in the field of precision drilling technology, Botek Przisionsbohrtechnik is an essential part of the project, both in carrying out the experiments and in validating the transfer learning. Empolis Information Management is responsible for the data preparation as well as the development of the ML models and ensuring the transferability. The analysis of machining mechanisms in drilling and milling using parametric models and machine learning is being carried out at the Chair of Production Engineering and Organisation FBK at the Technical University of Kaiserslautern. The tool manufacturer Paul Horn is responsible for carrying out and evaluating the milling tests and plays a major role in the data preparation. The Institute for Machine Tools IfW at the University of Stuttgart focuses on research into process pre-control and is responsible for the work at the interface between ML models and machine control. The project is scheduled to run until 31 May 2024.

* Mathias Schmidt is managing partner of K.-H. Mller Przisionswerkzeuge GmbH.

(ID:48097648)

Read more from the original source:

Production methods of tomorrow: More efficient tool use through artificial intelligence - ETMM Online

Related Posts