The purpose with Robust Engineering is to operate a system in such way that it mitigates incoming variation, in contrast to the old fashion way of improving quality by demanding higher precision and tighter tolerances of the inputs that by default makes a system more expensive. One of the issues that arise during robot welding of larger components is that incoming variations can differ substantially between individual parts. Gaps and angles between sheets can vary due to material variation, cutting or variation in fixturing. Geometrical variations, both due to the welding setup and due to the above described variations on incoming parts can thus lead to large deviations in final weld geometry causing increased cost of poor quality in terms of scrap, rework, overproduction (over-sized welds), waste of consumables and extra cost for inspection and monitoring as a result. Also, there is a risk that a bad weld reaches customers, which might lead to devastating consequences.
This project serves to demonstrate the potential of adaptive control of welding processes based on a combination of predictive models built from domain knowledge from offline experimental modelling, and machine learning from online data. With the addition of a learning component to weld process control it will be possible to achieve a robust weld result by automated adjustment of process parameters such as weld parameters and robot path, including torch angles and positioning, based on incoming part variations; thus achieving a robust weld result. By studying relations between weld parameters, robot positioning and resulting weld geometry it is also intended to demonstrate the possibility to predict inner weld quality (penetration depth) based solely on outer weld quality aspects when correlation is known.
This project will focus on MIG/MAG welding related to heavy vehicle applications such as haulers, cranes and forklifts. The results can however be applied to a broader range of welding industry.
Universitetslektor vid Chalmers, Industrial and Materials Science, Engineering Materials
Hisings Kärra, Sweden
Funding Chalmers participation during 2017–2018
Areas of Advance