Artificial Intelligence Planning of Failure Recovery Strategies in Discrete Manufacturing Automation

As discrete manufacturing tends to be small batch and customized, automated Production Systems (aPS) must be more flexible to adapt to the variety of products, which makes the aPS more complex and error-prone. To increase system efficiency and reduce downtime caused by manual intervention, strategies for automatic recovery are required. Currently, automatic recovery is fulfilled by portions of the control software that treat selected failures, which have been planned and implemented at design-time. To save engineering effort and treat unpredicted failures, recovery strategies should instead be generated automatically using established artificial intelligence planners. Consequently, this paper proposes a modularization of the functional control software into Control Primitives, from which generated strategies are composed. The same Control Primitives are the building blocks to manually implement the state machines of different operating modes of the aPS, Thus, no additional engineering effort is needed to prepare recoverability in the application development phase. In this paper, four approaches to model and implement PLC-executable Control Primitives are presented. Their viability is evaluated experimentally on a discrete manufacturing demonstrator machine by generating strategies for three use cases.