MATerial THicknEss Weldability (MATTHEW) from Inpro uses AI for spot weldability analysis
Testing the weldability of multi-sheet joints is an important technical prerequisite for the production of car bodies. So far, testing has mainly been carried out experimentally. A dedicated machine learning model is now used to predict the weldability of multi-sheet joints, thereby reducing experimental testing efforts and indirect factory costs significantly.
Challenge - Analyzing weldability for various material mixes with little manual effort
Welding is a key component when assembling a new car. It consists of joining together metal parts by heating the surfaces to the melting point and uniting them by pressing. Hence, welding requires the right mix of materials to ensure stability and longevity of not only the metal parts being welded, but also the car itself.
The material mix consists of different metal grades, sheet thicknesses, and coatings. The material mix determines the weldability of the respective metal parts. However, in today’s world, plants are facing an increasing number of new metal grades and sheet thicknesses. Therefore, there are more and more options in selecting different types of material parts. As a result, each new material mix must undergo weldability checks. For the Joining Technology Center, which performs weldability checks centrally for Volkswagen Passenger Cars and other selected Volkswagen brands, weldability checks have become more challenging and experimental tests are conducted. This is extensive and time-consuming and usually requires contracting third-party laboratories, incurring indirect factory costs.
Solution - Implement a machine learning model that predicts the weldability of new material mixes
The solution MATTEW is a core model based on machine learning, which consists of a web front-end. The machine learning model operates within the Digital Production Platform and can predict the weldability of new material mixes. In order to generate this result, the model links two sets of data. First, it utilizes an existing dataset containing physicochemical and technical properties and thicknesses of each material as well as other information, such as with or without coating. Second, the model links this existing dataset with the information about weldability (yes/no). As an output, the machine learning model produces a labelled dataset called Feature Vector. The Feature Vector contains all necessary information that is needed to train the machine learning model.
MATTHEW allows its users to import the initial list which is automatically converted into a material-thickness combination list. The machine learning model is then applied to this list. Therefore, the entire workflow, like uploading the initial list, converting datasets, and running the machine learning model, is controlled by MATTHEW in the Cloud using a web front.
Key benefits – Smooth workflow and cost-savings with help of automation
MATTHEW leads to a significant reduction of cost-intensive experimental testing by automating weldability checks, thereby reducing indirect factory costs.
MATTHEW - our new AI solution to automate weldability checks for all major Volkswagen car brands has significantly reduced cost-intensive testing efforts at our technology center and affiliated labs.