Strength Prognosis JbyF: Strength Prognosis for Joining by Forming

The aim of this project is to enable a fast and reliable prognosis of the mechanical strength of clinched connections and rivets (Self-Piercing Rivet, SPR) based on geometrical properties of the connections and basic material properties of the materials to be joined via data-based algorithms. Members of the target group are directly or indirectly involved in the development and / or implementation of innovative joining by forming technology. First of all, there are manufacturers and suppliers of mechanical connection technology who are constantly interacting with the manufacturing industry. This sector is increasingly resorting to numerical simulations - either in-house or via service providers. This implies that numerical simulation and advanced material identification service providers are also part of the development process. This project aims to create a digital ecosystem for the benefit of all major players in the innovation chain of mechanical connection technology in order to explore the possibilities of machine learning However, the vast majority of the user group is SME and does not have the resources to do this in-house . This project therefore has the ambition, through intense collaboration and a clearly defined pilot project, to i) increase the level of competition with regard to machine learning in connection design of 30 Flemish SMEs and to valorise them in a digital ecosystem, ii) to introduce 100 Flemish companies to the possibilities of machine learning in mechanical design.

Currently the design of mechanical joining processes like clinching and self-pierce riveting is hampered by a costly and complex experimental test campaign. Such approach is required to determine the most optimal process parameters such as the tool geometry and the punch stroke. A joint configuration is then iteratively varied based on experience until a suitable joint contour and strength is achieved. Fortunately, the latter process is nowadays highly supported by numerical simulations enabling to optimize the process parameters in a virtual environment. There is no doubt that these numerical simulations mitigate the burden of the experimental trial-and-error phase for optimizing the joining process. However, besides the joining process, the mechanical behaviour of the joint should also match the functional requirements of the mechanical design at hand. To this end, mechanical experiments are conducted to determine the mechanical strength of the joint under different loading conditions. In structural applications, these experiments are of crucial importance, yet increase the overall design cycle of the joint. A reliable estimation of the joint strength is crucial in an early stage of the joint design. This can be done via experiments and numerical simulations. The bottleneck, however, is that both the experimental route and the numerical route are time-consuming. In this project, we aim at reducing the development cycle of mechanical joining technologies by targeting a strength prognosis based on data-based and analytical models.

Two prediction methods, namely machine learning and analytical formulas, are investigated for strength prediction of self-pierce riveted or clinched joints. For both methods, we aim at an instant prediction of the mechanical strength so that a significant speedup factor can be achieved in the overall design process of the joint. The machine learning approach relies on an extensive experimental and numerical database that was generated in this project. The experimental database accounts for joining of steel and aluminium sheets with tensile strengths ranging from 200 MPa up to 1000 MPa. The database is then be used for the training and the evaluation of different strength prediction models.

The analytical strength prediction starts from an in-depth literature survey. A summary of the available formula’s is given along with the underlying assumptions. The analytical approaches are validated against the extensive experimental database enabling to select the most appropriate analytical prediction for each strength test. Moreover, the study revealed some shortcomings of currently applied analytical predictors. Consequently, some modifications to increase the predictive accuracy of analytical methods are proposed in this project.

Finally, in the framework of the project, a web application was developed enabling to instantly predict the strength of mechanical joints via machine learning and analytical methods. The user can input parameters such as material properties and process parameters. The developed prediction models are implemented in the web application and instantly report the quasi-static mechanical strength of the joint. Moreover, the webapp enables the user to profoundly understand the machine learning approach. Interested parties can request access to the "machine learning method" section enabling to develop a model with new data.

The web application can be reached via Strength Prognosis JbyF. 

The project results were published and can be found here.

Please contact Sam Coppieters (sam.coppieters@kuleuven.be) for more information.