Using statistically designed experiments for safety system optimization

Abstract

This paper describes the method of statistically designed experiments (SDEs) which is used as a structured method to investigate the best setting for a number of decision variables in a system design problem. Traditionally, in the design of safety-critical systems, a trial and error type approach is undertaken to achieve a final system that meets the design objectives. This approach can be time consuming, and often only an adequate design is found rather than the optimal design for the available resources. Optimal use of resources should be imperative when possible lives are at risk. To demonstrate the pracaticality of this new structured approach for optimizing a safety system design, a high-integrity safety system has been used. Each design is analysed using the binary decision diagram analysis technique to establish the system unavailability, which is penalized if the system constraints are exceeded. System constraints indicate the limitations on the resources that can be utilized. The SDE approach highlights good and bad settings for possible design variables. This knowledge can then be used by more sophisticated search techniques. The latter part of this paper analyses the results from the best design generated using the SDE, for further optimization using localized optimization approaches.