Citation

Aly S, Tyrychtr J, Kvasnicka R, Vrana I. Safety Sci. 2021; 140: 105292.

Copyright

(Copyright © 2021, Elsevier Publishing)

DOI

10.1016/j.ssci.2021.105292

PMID

unavailable

Abstract

The practice of developing product or system safety standards is currently conducted with high level of subjectivity, which affects the reliability of development outcomes. The decision to adopt specific safety requirements into the standard is usually based on the knowledge, intuition and experience of the technical expert involved in developing the body of the safety standard made up of those requirements or specifications. This is usually carried out without analyzing risk attributes or factors addressed by each safety requirement constituting a safety standard. This can lead to inaccuracy and lack of reliability due to either under-estimation or over-estimation of risks involved. This research presents unprecedented, novel, structured and objective methodology, that aims to improve the practice of developing safety standards. In order to accomplish the above end, the proposed methodology acts to prioritize the system' safety requirements of each system's design elements based on a set of commonly identified risk factors they address. We have investigated the currently existing risk prioritization and multi-criterion decision making techniques, and found that machine learning data clustering approaches can be adequate especially in case of relatively large number of assessed decision alternatives (i.e., the assessed safety requirements). We have elaborated the merits of the proposed clustering-based methodology over existing prominent ones. The proposed methodology starts with analyzing product or system safety to identify common risk factors. A numerical scale is used to enable objective quantification of risk factors' values for each safety requirement by multiple experts. The obtained experts' numerical assessments of each safety requirement are then averaged to represent the raw data set containing risk profiles of each safety requirement assessed. This risk factors data set is then used by fuzzy c-means clustering algorithm, to organize them into groups of different level of priorities, so as to prioritize their corresponding safety requirements. We have applied the proposed methodology on a real case study of developing school bus safety standard. The main benefit of the proposed methodology includes its capability to support the safety standards development practitioners and policy makers, as well as technical experts in objectively and systematically carry out the critical decision of adoption, exclusion and update of the safety requirements into the developed safety standard. Additionally, it can efficiently guide safety engineers and inspectors to undertake more reliable and structured risk assessment process. Over and above, the implementation of the proposed methodology is guaranteed to result in more reliable, accurate and risk-informed safety standard.

Language: en

Keywords

Clustering algorithm; Fuzzy c-means; Risk assessment; Safety requirements assessment and prioritization; Safety standards development