A behaviour-based approach to the design of safety-critical systems

Software-based systems usually have complex functionality, and are required to meet very high integrity and reliability requirements. Testing a software-based system to determine whether it meets such requirements is impractical, requiring test runs whose accumulated time is much greater than 10⁹ hours, in order to demonstrate the above reliability requirement. Software does not fail in a random manner; if a program produces an incorrect output (assuming that no hardware failure corrupted the program, which is a different scenario to a software failure), then such an output is inherent in its structure, i.e. the software contains a design error. This makes the use of traditional analysis methods unsuitable, because they are generally suited to assessing random failures in hardware. We present an approach which shows promise in enabling software to be assessed directly for safety. The approach is centred on the use of certain technologies, drawn from the artificial intelligence domain, namely decision trees and behaviour based architectures (e.g. subsumption architecture). The paper describes the features of these technologies, which will be useful in the design and construction of safety critical systems