Clustered Problems and Machine Learning Methodologies: A New Approach

The future of production is smart and autonomous and so should the technologies that allow it. The application of machine learning (ML) in the manufacturing sector has been designated as one of the key players in the digital transformation. In general, the tendency of most ML application fields goes into more complex algorithms and deeper architectures. However, it is not the only available approach and often times not the most suitable for all use cases in the manufacturing industry where environments and products are highly standard. ML has the potential to solve complex problems by finding and learning relationships that humans cannot identify. Our hypothesis is that these relationships are invariant and exist inside specific tasks but also between a wide catalogue of similar, although not yet known, problems. An algorithm that finds and solves these invariances at a general and abstract level creates a general solution that can be directly deployed into new use cases leading to an increase in efficiency by reducing the required effort, time, and knowledge and allowing to leverage ML. Through three preliminary use cases, we demonstrate the potential of applying ML at a general and abstract level to increase the flexibility and reduce the model, task, and dataset dependency of ML solutions. This approach facilitates the conversion of automated environments into autonomous environments.