Despite its great success, machine learning can have its limits when dealing with insufficient training data. A potential solution is the additional integration of prior knowledge into the training process which leads to the notion of informed machine learning. In this paper, we present a structured overview of various approaches in this field. We provide a definition and propose a concept for informed machine learning which illustrates its building blocks and distinguishes it from conventional machine learning. We introduce a taxonomy that serves as a classification framework for informed machine learning approaches. It considers the source of knowledge, its representation, and its integration into the machine learning pipeline. Based on this taxonomy, we survey related research and describe how different knowledge representations such as algebraic equations, logic rules, or simulation results can be used in learning systems. This evaluation of numerous papers on the basis of our taxonomy uncovers key methods in the field of informed machine learning.

The automotive industry is increasingly focusing on product customization. The concept of Modular Production addresses this issue by providing more flexibility in production with Automated Guided Vehicles transporting products between modular workstations. The added complexity of Modular Production Control calls for approaches that can handle the scheduling complexity while also minimizing production costs. As a result, literature has focused on two promising approaches: Deep Reinforcement Learning and Multi-Agent Systems. Both approaches have their advantages. Especially in complex, large-scale production environments with random breakdowns, those two fields have been seldomly combined, though. As a result, this article aims to fill that research gap by introducing a Deep Reinforcement Learning Multi-Agent System approach for Modular Production Control. We introduce a reward design incentivizing agents to achieve maximal throughput. In addition, we show that the method learns optimal behavior even in a large-scale production environment with random machine breakdowns. Lastly, we compare the Multi-Agent System to a single-agent implementation of the Deep Reinforcement Learning approach and conclude that the Multi-Agent Deep Reinforcement Learning method learns and solves the Modular Production Control problem with the same solution quality as the single agent. Hence, the approach allows to foster MAS benefits such as robustness without losses in the solution quality.

In this paper, we describe the combination of machine learning and simulation towards a hybrid modelling approach. Such a combination of data-based and knowledge-based modelling is motivated by applications that are partly based on causal relationships, while other effects result from hidden dependencies that are represented in huge amounts of data. Our aim is to bridge the knowledge gap between the two individual communities from machine learning and simulation to promote the development of hybrid systems. We present a conceptual framework that helps to identify potential combined approaches and employ it to give a structured overview of different types of combinations using exemplary approaches of simulation-assisted machine learning and machine-learning assisted simulation. We also discuss an advanced pairing in the context of Industry 4.0 where we see particular further potential for hybrid systems.

In industry, individualization leads to a slow replacement of assembly line production with more flexible modular assembly systems. In modular systems, each product can be completed on multiple routes through a grid of modular workstations, where transportation is handled by automated guided vehicles (AGV). In order to benefit from this routing flexibility and to react on disturbances in the system, new robust control approaches are crucial. While optimizing the production flow globally is limited by computing power, this work presents a decentralized control approach that reduces complexity by dividing the problem into sub-problems: A job release agent releases jobs at certain points in time according to the system's inventory level. Each job in the system is linked to a job routing agent regularly choosing the optimal route out of the options given by the product's flexibility. Every modular station is represented by a workstation agent optimizing the workstation's schedule. Lastly, a vehicle agent assigns transports optimally to the AGVs and coordinates them accordingly. An evaluation example emphasized the decentralized approach as a valid way for real-time robust control solutions, where the makespan was about five percent away from the static optimum.

Addressing the need for more flexible modular assembly systems due to an increasing number of product variants, industry utilizes the capabilities of cyber-physical production systems, where automated guided vehicles handle the transport of materials. Such systems require complex control algorithms managing the given flexibility and coordinating the cyber-physical entities. Control algorithms can be organized either in a centralized or decentralized manner. For the development and the performance evaluation of both types of algorithms, no standardized framework exists using the same boundary conditions. Following the principles of virtual commissioning, this work presents a framework composed of three elements: a simulation model representing a modular assembly system, a multi-agent system incorporating centralized or decentralized control logic, and an interface for data-exchange. The framework has been successfully validated implementing a centralized approach, where the system followed a global schedule and a decentralized approach, where a bidding-based agent-system controlled the production flow.

An increasing number of product variants caused by individualization drives industry to develop flexible modular assembly systems. In such systems, automated guided vehicles handle the material transport between decoupled workstations. This enables jobs to be flexibly routed through production, utilizing order and operation flexibility. Order flexibility denotes a partially flexible order of a job's operations, whereas operation flexibility is given, if at least one operation of a job is allocatable to more than one workstation. Managing the given flexibility and coordinating the automated guided vehicles in real production environments requires robust control algorithms. Exploiting the potential of Industry 4.0, this work presents a negotiation-based agent approach by decentralizing the problem: A job release agent defines the points in time and the product types of released jobs. Job routing agents negotiate with workstation agents to schedule their operations by minimizing makespan. Finally, a vehicle agent organizes the transports from workstation to workstation. The agent system has been successfully evaluated and achieved better results than earlier implementations.