Machine manufacturers and plant operators must continuously adapt to global competition. In addition to the existing challenges posed by disrupted supply chains, market conditions and energy costs, the focus is on two key requirements: customer requirements are increasing in complexity - both in terms of the machines themselves and in terms of IT integration - and the demand for smaller batch sizes requires machines to be quickly converted for different product variants. At the same time, these market changes are opening up exciting opportunities to improve one's own market position, as well as potential for reducing costs, increasing production efficiency and making systems more flexible and reconfigurable.

New production concepts

Some of the new requirements break with the previous focus on high cycle rates. They require machine builders and plant operators to re-evaluate existing products and develop new production concepts:

• Resilient value networks and supply chains
• Digital twins for materials, processes, production networks and the product as a whole
• Automated product configuration through AI systems
• Self-organizing process route planning

Process mining methods are used to analyse existing processes in order to identify optimization potential in the production lines. The results reveal deviations and bottlenecks by comparing the planned process with the actual workflow on the store floor. It becomes clear that, although rigid line networks are efficient for fixed product groups, they quickly reach their limits when it comes to variable product requirements. Ideally, the line should be individually adapted for each variable product group. The use of autonomous mobile machine tools and intelligent software agents enables automated adaptation of the production line to the specific product requirements. The software agents provide the machines with all the information they need to automatically group themselves into a new production line.

Software agents and agent systems

A software agent fulfills basic core functions: It is aware of its environment and can grasp it, it can interact with the environment and this interaction enables the agent to perform its task optimally. An agent is more than a digital twin, as it acts with foresight, consciously navigates through its environment and achieves goals independently. In this concept, an agent always represents a component of the real system, be it a machine, a workpiece or an order. In order to support an entire production hall, many agents are required that communicate and cooperate with each other. This creates a decentralized multi-agent system (MAS) that supports the entire production concept.
The implementation of the tasks of the multi-agent system (MAS) on real components requires both a connection to the environment by means of sensors and actuators, as well as networked communication via any communication protocols and algorithms for target determination.

Recently, approaches have been developed to automatically adapt the entire line structure to the product to be produced using a MAS. In these systems, the agents control the assembly of the line using Automated Guided Vehicle (AGV)-based machines, which can be coupled as required and carry out various production and process steps. Control is no longer centralized, but decentralized by intelligent agents on the store floor. This dissolves the traditional hierarchy of the production line and leads to a heterarchy of various equivalent software agents and hardware components. The resulting flexible and independent hardware modules can be used in a variety of ways. The autonomous, decentralized system also complies with RAMI 4.0 standards.

AI for finding the agent's goal

In mechanical engineering, AI is generally used in the context of machine learning, either to predict machine states in advance or to recognize patterns as part of the manufacturing process. In the environment outlined above, AI has a new task: to optimize the product-controlled structure of the production line by determining the optimal use of space on the store floor in order to operate as many production lines as possible simultaneously and to carry out job planning taking into account the use of space in order to meet delivery deadlines and other production specifications. AI can be trained for both requirements: It receives the current status of the environment and planned targets via the MAS, whereupon it can respond with production forecasts. It uses this knowledge to suggest to the agent in the production line the best current route to achieve the target. Depending on the agent's task, this can relate to the line structure, processing in the process, scheduling or any other activity.
The advantage of the overall system is that many steps can be carried out automatically without the need for manual intervention in production planning. The transparency of production is maintained, as is the possibility of manual intervention.