Search

Artificial Intelligence (AI) can enable the development of next-generation autonomous safety-critical systems in which Machine Learning (ML) algorithms learn optimized and safe solutions. AI can also support and assist human safety engineers in developing safety-critical systems. However, reconciling both cutting-edge and state-of-the-art AI technology with safety engineering processes and safety standards is an open challenge that must be addressed before AI can be fully embraced in safety-critical systems. Many works already address this challenge, resulting in a vast and fragmented literature. Focusing on the industrial and transportation domains, this survey structures and analyzes challenges, techniques, and methods for developing AI-based safety-critical systems, from traditional functional safety systems to autonomous systems. AI trustworthiness spans several dimensions, such as engineering, ethics and legal, and this survey focuses on the safety engineering dimension., Peer Reviewed, Postprint (published version)

Artificial Intelligence (AI) can enable the development of next-generation autonomous safety-critical systems in which Machine Learning (ML) algorithms learn optimized and safe solutions. AI can also support and assist human safety engineers in developing safety-critical systems. However, reconciling both cutting-edge and state-of-the-art AI technology with safety engineering processes and safety standards is an open challenge that must be addressed before AI can be fully embraced in safety-critical systems. Many works already address this challenge, resulting in a vast and fragmented literature. Focusing on the industrial and transportation domains, this survey structures and analyzes challenges, techniques, and methods for developing AI-based safety-critical systems, from traditional functional safety systems to autonomous systems. AI trustworthiness spans several dimensions, such as engineering, ethics and legal, and this survey focuses on the safety engineering dimension., Validerad;2024;Nivå 2;2024-07-09 (joosat);Full text license: CC BY

The development of dependable and intelligent embedded systems progresses. However, integrating complex software stacks, machine learning solutions, and high-performance computing devices amplifies the functional and temporal uncertainty sources engineers must manage when developing dependable systems and software., Peer Reviewed, Postprint (author's final draft)

The ability of CNNs to efficiently and accurately perform complex functions, such as object detection, has fostered their adoption in safety-related autonomous systems. These algorithms require high computational performance platforms that exploit high levels of parallelism. The detection, control and mitigation of random errors in these underlying high computational platforms become a must according to functional safety standards. In this paper, we propose protecting, with a catalog of diagnostic techniques, the most computationally expensive operation of the CNNs, the matrix multiplication. However, this protection entails a performance penalty, and the complete CNN protection may be unaffordable for those systems operating with strict real-time constraints. This paper proposes a three-stage methodology to selectively protect CNN layers to achieve the required diagnostic coverage and performance trade-off: i) sensitivity analysis to misclassification per CNN layers using a statistical fault injection campaign, ii) layer-by- layer performance impact and diagnostic coverage analysis, and iii) selective layer protection. Furthermore, we propose a strategy to effectively compute the achievable diagnostic coverage of large matrices implemented on GPUs. Finally, we apply the proposed methodology and strategy in Tiny YOLO-v3, an object detector based on CNNs., Ikerlan authors have received funding from Elkartek grant project KK-2021/00123 of the Basque government. BSC au- thors have been partially supported by the Spanish Ministry of Science and Innovation under grant PID2019-107255GB- C21/AEI/10.13039/501100011033, Postprint (author's final draft)

An increasing number of critical functionalities integrated in embedded critical systems rely on deep learning (DL) technology. This article summarizes certain key aspects of DL’s intrinsic stochastic and training-data-dependent nature that are at odds with current domain-specific functional safety standards. We exemplify how redundancy and diversity of neural networks can help to reconcile DL technology and functional safety requirements., The research leading to these results has received funding from the European Research Council (ERC) grant agreement No. 772773 (SuPerCom), the Horizon Europe Programme under the SAFEXPLAIN Project (www.safexplain.eu), grant agreement num.101069595, and the Spanish Ministry of Science and Innovation under grant PID2019-107255GBC21/AEI/10.13039/501100011033., Peer Reviewed, Postprint (author's final draft)

Multi-core devices are envisioned to support the development of next-generation safety-critical systems, enabling the on-chip integration of functions of different criticality. This integration provides multiple system-level potential benefits such as cost, size, power, and weight reduction. However, safety certification becomes a challenge and several fundamental safety technical requirements must be addressed, such as temporal and spatial independence, reliability, and diagnostic coverage. This survey provides a categorization and overview at different device abstraction levels (nanoscale, component, and device) of selected key research contributions that support the compliance with these fundamental safety requirements., This work has been partially supported by the Spanish Ministry of Economy and Competitiveness under grant TIN2015-65316-P, Basque Government under grant KK-2019-00035 and the HiPEAC Network of Excellence. The Spanish Ministry of Economy and Competitiveness has also partially supported Jaume Abella under Ramon y Cajal postdoctoral fellowship (RYC-2013-14717)., Peer Reviewed, Postprint (author's final draft)