25 Mar 2020 | DataProphet| • General News
DataProphet has reduced spot weld defects in the body shop of a mature vehicle assembly plant by 75%. We share key takeaways and tips from our spot weld experience and other real-world implementations of AI to optimize manufacturing control parameters.
In our experience, the most successful process optimization solutions are both holistic and proactive. Holistic process models capture the complexities of a system arising out of the extreme interdependence in a chain of processes. This includes the interactions between process parameters, higher-order effects that changes in one part of the system have on other parts, and physical constraints within which the process must operate.
The DataProphet approach minimizes the cost and complexity, and the potential unintended impact of an AI augmentation. In our vehicle spot welding system, as in many other deployments, we find that by empowering specialists with an Expert Execution System (EES), the factory technicians embraced the system as a powerful addition to their toolset; receiving the attribution and praise for the performance improvement.
In our vehicle assembly case study, we took this transfer of expertise a step further: our client’s typical body shop contains dozens of clusters of robots, or “stations”. Each station conducted a weld at dozens of predefined locations on the vehicle. Within one technology, such as spot welding, these welds share a common set of controllable parameters, but each was uniquely specified in terms of their set-points. This is because each weld location is unique in important ways, such as the type and thickness of the materials being joined at that location. Given this difference between weld locations, and the discrete nature and relative uniformity of the succession of welds at each weld location, we took the natural approach of modeling each weld location as though it were its own mini-factory, and of modeling each weld as a discrete unit of production.
Factory processes are not stationary. Once-off and even regular data science exercises, resulting in a static set of AI-enabled insights or a static AI model, will fail to address this fact. In order to remain relevant and valuable, an AI model needs to continuously optimize the factory process(es) as upstream conditions change, which in turn requires the AI model to be kept up-to-date and to be made available “on-tap” to factory technicians.
Read the full paper here to learn how you can optimize your plant control plans.
22 Sep 2022 Nicol Ritchie
The previous article in our change management series made a case for embedding algorithmic business ...
17 Aug 2022 Nicol Ritchie
Manufacturing change management wisdom is a give-and-take between technology and people. However, high-impact AI solutions must ...
26 Jul 2022 Nicol Ritchie
Acting on data is now key to affecting last-mile optimization of production. However, reskilling plant ...
8 Jul 2022 Nicol Ritchie
In a wide range of manufacturing contexts, pressure is building to empower manufacturing teams to ...