Challenges and solutions for data-driven process optimization
Machine learning can be used to predict product quality, machine downtimes or process conditions, in order to then take the right measures at the right time. Data scientists or process experts are responsible to interpret the analysis result. However, current problems interpreting these results are the scope for interpretation, inherent uncertainties in the models or a lack of knowledge about the correct reaction steps. The use of fully or semi-autonomous systems offers the possibility to exploit optimization potentials in production and to overcome the challenges mentioned above. Even companies that lack the necessary specialists could optimize individual processes or process chains with the help of machine learning models.
The future of process optimization
Fraunhofer IPT conducts research on the implementation of fully and partially autonomous systems for the deployment of machine learning models in production. Multi-agent systems and service-oriented architectures enable the integration of an AutoML pipeline. AutoML refers to the process of automating some or all parts of the machine learning pipeline with the goal of reducing the user's workload.
By using optimizers, process parameters can be adjusted based on predictions of the machine learning models in order to achieve optimal quality of the final products and to control production systems completely autonomously. It must be ensured that the result of the machine learning model cannot endanger plant and worker safety. Good traceability of the models makes the results easier for process experts to interpret. The aim is to certify the models and applications of Machine Learning in order to cover a broad industrial application.
Our range of services
- Process optimization in production
- Deployment of machine learning algorithms in production
- Feasibility studies