Safety-critical Applications Research Articles

Explicitly Perceiving and Preserving the Local Geometric Structures for 3D Point Cloud Attack

Deep learning models for point clouds have shown to be vulnerable to adversarial attacks, which have received increasing attention in various safety-critical applications such as autonomous driving, robotics, and surveillance. Existing 3D attack methods generally employ global distance losses to implicitly constrain the point-wise perturbations for optimization. However, these simple losses are quite difficult to accurately measure and restrict the proper 3D geometry as point clouds are highly structured. Although few recent works try to exploit additional shape-aware surface knowledge to globally constrain the point position, they still fail to preserve the detailed point-to-point geometric dependency in different local regions. To this end, in this paper, we propose a novel Multi-grained Geometry-aware Attack (MGA), which explicitly captures the local topology characteristics in different 3D regions for adversarial constraint. Specifically, we first develop multi-scale spectral local filter banks adapting to different 3D object shapes to explore potential geometric structures in local regions. Considering that objects may contain complex geometries, we then extend each filter bank into multi-layer ones to gradually capture the topology contexts of the same region in a coarse-to-fine manner. Hence, the focused local geometric structures will be highlighted in the coefficients calculated by the filtering process. At last, by restricting these coefficients between benign and adversarial samples, our MGA is able to properly measure and preserve the detailed geometry contexts in the whole 3D object with trivial perturbations. Extensive experiments demonstrate that our attack can achieve superior performance on various 3D classification models, with satisfying adversarial imperceptibility and strong resistance to different defense methods.

DeepSaDe: Learning Neural Networks That Guarantee Domain Constraint Satisfaction

As machine learning models, specifically neural networks, are becoming increasingly popular, there are concerns regarding their trustworthiness, specially in safety-critical applications, e.g. actions of an autonomous vehicle must be safe. There are approaches that can train neural networks where such domain requirements are enforced as constraints, but they either cannot guarantee that the constraint will be satisfied by all possible predictions (even on unseen data) or they are limited in the type of constraints that can be enforced. In this paper, we present an approach to train neural networks which can enforce a wide variety of constraints and guarantee that the constraint is satisfied by all possible predictions. The approach builds on earlier work where learning linear models is formulated as a constraint satisfaction problem (CSP). To make this idea applicable to neural networks, two crucial new elements are added: constraint propagation over the network layers, and weight updates based on a mix of gradient descent and CSP solving. Evaluation on various machine learning tasks demonstrates that our approach is flexible enough to enforce a wide variety of domain constraints and is able to guarantee them in neural networks.

VeriCompress: A Tool to Streamline the Synthesis of Verified Robust Compressed Neural Networks from Scratch

AI's widespread integration has led to neural networks (NN) deployment on edge and similar limited-resource platforms for safety-critical scenarios. Yet, NN's fragility raises concerns about reliable inference. Moreover, constrained platforms demand compact networks. This study introduces VeriCompress, a tool that automates the search and training of compressed models with robustness guarantees. These models are well-suited for safety-critical applications and adhere to predefined architecture and size limitations, making them deployable on resource-restricted platforms. The method trains models 2-3 times faster than the state-of-the-art approaches, surpassing them by average accuracy and robustness gains of 15.1 and 9.8 percentage points, respectively. When deployed on a resource-restricted generic platform, these models require 5-8 times less memory and 2-4 times less inference time than models used in verified robustness literature. Our comprehensive evaluation across various model architectures and datasets, including MNIST, CIFAR, SVHN, and a relevant pedestrian detection dataset, showcases VeriCompress's capacity to identify compressed verified robust models with reduced computation overhead compared to current standards. This underscores its potential as a valuable tool for end users, such as developers of safety-critical applications on edge or Internet of Things platforms, empowering them to create suitable models for safety-critical, resource-constrained platforms in their respective domains.

Rethinking Robustness of Model Attributions

For machine learning models to be reliable and trustworthy, their decisions must be interpretable. As these models find increasing use in safety-critical applications, it is important that not just the model predictions but also their explanations (as feature attributions) be robust to small human-imperceptible input perturbations. Recent works have shown that many attribution methods are fragile and have proposed improvements in either these methods or the model training. We observe two main causes for fragile attributions: first, the existing metrics of robustness (e.g., top-k intersection) overpenalize even reasonable local shifts in attribution, thereby making random perturbations to appear as a strong attack, and second, the attribution can be concentrated in a small region even when there are multiple important parts in an image. To rectify this, we propose simple ways to strengthen existing metrics and attribution methods that incorporate locality of pixels in robustness metrics and diversity of pixel locations in attributions. Towards the role of model training in attributional robustness, we empirically observe that adversarially trained models have more robust attributions on smaller datasets, however, this advantage disappears in larger datasets. Code is made available at https://github.com/ksandeshk/LENS.

Towards Trustworthy Deep Learning

Deep neural networks (DNNs) have achieved unprecedented success across many scientific and engineering fields in the last decades. Despite its empirical success, unfortunately, recent studies have shown that there are various failure modes and blindspots in DNN models which may result in unexpected serious failures and potential harms, e.g. the existence of adversarial examples and small perturbations. This is not acceptable especially for safety critical and high stakes applications in the real-world, including healthcare, self-driving cars, aircraft control systems, hiring and malware detection protocols. Moreover, it has been challenging to understand why and when DNNs will fail due to their complicated structures and black-box behaviors. Lacking interpretability is one critical issue that may seriously hinder the deployment of DNNs in high-stake applications, which need interpretability to trust the prediction, to understand potential failures, and to be able to mitigate harms and eliminate biases in the model. To make DNNs trustworthy and reliable for deployment, it is necessary and urgent to develop methods and tools that can (i) quantify and improve their robustness against adversarial and natural perturbations, and (ii) understand their underlying behaviors and further correct errors to prevent injuries and damages. These are the important first steps to enable Trustworthy AI and Trustworthy Machine Learning. In this talk, I will survey a series of research efforts in my lab contributed to tackling the grand challenges in (i) and (ii). In the first part of my talk, I will overview our research effort in Robust Machine Learning since 2017, where we have proposed the first attack-agnostic robustness evaluation metric, the first efficient robustness certification algorithms for various types of perturbations, and efficient robust learning algorithms across supervised learning to deep reinforcement learning. In the second part of my talk, I will survey a series of exciting results in my lab on accelerating interpretable machine learning and explainable AI. Specifically, I will show how we could bring interpretability into deep learning by leveraging recent advances in multi-modal models. I'll present recent works in our group on automatically dissecting neural networks with open vocabulary concepts, designing interpretable neural networks without concept labels, and briefly overview our recent efforts on demystifying black-box DNN training process, automated neuron explanations for Large Language Models and the first robustness evaluation of a family of neuron-level interpretation techniques.

Understanding Likelihood of Normalizing Flow and Image Complexity through the Lens of Out-of-Distribution Detection

Out-of-distribution (OOD) detection is crucial to safety-critical machine learning applications and has been extensively studied. While recent studies have predominantly focused on classifier-based methods, research on deep generative model (DGM)-based methods have lagged relatively. This disparity may be attributed to a perplexing phenomenon: DGMs often assign higher likelihoods to unknown OOD inputs than to their known training data. This paper focuses on explaining the underlying mechanism of this phenomenon. We propose a hypothesis that less complex images concentrate in high-density regions in the latent space, resulting in a higher likelihood assignment in the Normalizing Flow (NF). We experimentally demonstrate its validity for five NF architectures, concluding that their likelihood is untrustworthy. Additionally, we show that this problem can be alleviated by treating image complexity as an independent variable. Finally, we provide evidence of the potential applicability of our hypothesis in another DGM, PixelCNN++.

Robust Uncertainty Quantification Using Conformalised Monte Carlo Prediction

Deploying deep learning models in safety-critical applications remains a very challenging task, mandating the provision of assurances for the dependable operation of these models. Uncertainty quantification (UQ) methods estimate the model’s confidence per prediction, informing decision-making by considering the effect of randomness and model misspecification. Despite the advances of state-of-the-art UQ methods, they are computationally expensive or produce conservative prediction sets/intervals. We introduce MC-CP, a novel hybrid UQ method that combines a new adaptive Monte Carlo (MC) dropout method with conformal prediction (CP). MC-CP adaptively modulates the traditional MC dropout at runtime to save memory and computation resources, enabling predictions to be consumed by CP, yielding robust prediction sets/intervals. Throughout comprehensive experiments, we show that MC-CP delivers significant improvements over comparable UQ methods, like MC dropout, RAPS and CQR, both in classification and regression benchmarks. MC-CP can be easily added to existing models, making its deployment simple. The MC-CP code and replication package is available at https://github.com/team-daniel/MC-CP.

Sequentially Triggering “Time-Bomb” Trojan into Hardware Wired Microsequencer

This paper presents an analysis of a sequential Triggered “Time Bomb” hardware Trojan (HT). Major security concerns have been rising up since malicious modification of hardware or fabrication IC can lead to an altered functional behavior, potentially with disastrous consequences in safety-critical applications. Due to the stealthy nature of the hardware Trojans the conventional design and time verification and post manufacturing testing cannot be readily extended. There are a large number of possible instances and operating modes for the hardware Trojans in a digital system. Since the hardware Trojan insertion can modify the functionality of the digital integrated circuit (IC), alter its behavior, generate denial of service (DoS), the HT threats should be analyzed with maximum importance through the entire lifecycle of the IC.

Robust wear performance of graphene-reinforced high entropy alloy composites

The expected excellent tribological performance of high-entropy alloys (HEAs) is rarely achieved due to the absence of self-lubricating ability on the worn surface. In particular, the situation is more challenging in HEAs with face-centered cubic structure because of the inferior yield strength when used in dry sliding conditions. Unlike conventional soft-solid lubricants, which deteriorate the mechanical properties of HEA matrix, in this study we demonstrate in a protocol model alloy that the proper incorporation of few-layer graphene as reinforcement in CoCrFeNiMn can lead to both surface strengthening and lubrication. This is achieved by a partially chemical interface reaction between graphene and HEA via a tunable fabrication process, resulting in a reduction of wear rate and friction coefficient of 86.03% and 23.87%, respectively. As a result of the interfacial in-situ carbide strengthening together with the self-lubrication of graphene, this tailored composite structure also exhibits superior tribological properties over most HEA-based self-lubrication composites. Besides, the subsurface structure evolution and deformation mechanisms that influence the wear resistance are systematically clarified through microscopic exploration and atomistic simulation. The present study presents an effective strategy for the development of innovative HEA composites suitable for safety-critical applications, which overcomes the inherent compromise and achieves exceptional tribological performance, i.e., self-lubricating and anti-wear.

Advanced Data Science Applications in Vehicles: A Comprehensive Review

Purpose: This research paper aims to explore the transformative impact of data science applications within the automotive industry, with a focus on the evolution of vehicles into intelligent, connected, and autonomous entities. Through an extensive literature review and analysis of recent developments, the paper seeks to provide a comprehensive understanding of the current state and future prospects of data science in vehicles.
 Methodology: Advancements in sensor technologies, such as LiDAR, HD maps, radars, location tracking and cameras, are discussed, highlighting their role in data collection for applications like advanced driver assistance systems (ADAS) and autonomous driving. Relevant citations are provided. This subsection covers the collection and transmission of real-time data through telematics systems, showcasing their importance for predictive maintenance, fleet management, and personalized insurance programs. Citations support the presented information. The deployment of edge computing for real-time data processing in vehicles is discussed, emphasizing its significance for safety-critical applications like collision avoidance. Citations are provided to support the information. This section explores the application of machine learning algorithms to predict vehicle failures, optimize fuel efficiency, and analyze driver behavior. The importance of leveraging historical data to create accurate models is highlighted with supporting citations.
 Findings: The pivotal role of data science in enabling autonomous vehicles to navigate complex environments is discussed, emphasizing the use of machine learning models for real-time decision-making. Citations support the presented information. This subsection explores the integration of data science in traffic management systems, covering dynamic traffic signal control, congestion prediction, and route optimization. Citations support the findings related to traffic management applications. The paper discusses challenges associated with widespread implementation, including data privacy concerns, cybersecurity risks, and the need for standardized communication protocols. Additionally, potential future directions are outlined, such as the integration of blockchain for secure data sharing and the development of advanced human-machine interfaces.
 Unique Contribution to Theory, Practice and Policy: This research paper provides a well-rounded contribution by seamlessly integrating theoretical concepts, practical applications, and policy considerations in the realm of advanced data science applications in vehicles.

Understanding Cu Sintering and Its Role on Corrosion Behaviour for High-Temperature Microelectronic Application

There has been a significant rise in interest from academic and industry communities on copper nano and micro-particles-based sinter pastes for harsh environment die attach. However, sintering these particles is a complex process which is affected by many parameters, such as (i) particles size, shape, and distribution, (ii) sintering temperature, pressure, environment, and (iii) organic compounds used for the paste formulation. In literature, various research groups demonstrated that sintered copper layer can achieve good electrical conductivity and high strength. In our previous research [1, 2], we explained that microscale etched flakes are also capable of shear strength >30Mpa while sintering with 10 MPa bonding pressure at 275°C for just 1 minute with negligible paste cost (in comparison to silver paste). However, very limited knowledge exists on literature about corrosion behaviour in sintered copper interconnects. This, from experience in the industry with silver sintering has proven to be critical to applications. For examples in high power LED headlamp applications, minor exposure to residual Sulphur emerging from vulcanized rubber seals has been observed to cause corrosion in silver sintered interconnects. Therefore, corrosion and its impact on performance and reliability must be considered, especially for safety critical applications. For this reason, we have studied the relationship between copper sintered samples and corresponding corrosion behaviours and compared again to pure copper (conventionally produced). The corrosion resistance of the samples in salt water was assessed via various electrochemical analysis techniques and correlated with sintered microstructure. Sample produced better corrosion resistance when fabricated with higher density (less porosity). However, in case of the microscale copper flakes-based interconnects, a pitting behaviour was observed. The sintered copper paste samples showed to be nobler than commercial copper. In addition, copper oxide and its role in accelerating/preventing oxidation is also studied.

Gate-Oxide Degradation Monitoring of SiC MOSFETs Based on Transfer Characteristic With Temperature Compensation

Gate-oxide degradation is a concern in SiC MOSFETs especially in safety critical applications such as aerospace and electric vehicles (EV). To address this concern, this paper presents an accurate gate-oxide degradation monitoring solution based on the SiC devices’ transfer characteristics. Specifically, a plug-in circuit for gate driver is proposed to extract the transconductance ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">g_m</i> ) and threshold voltage ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V_th</i> ) values from transfer characteristic. By using the threshold voltage as gate-oxide aging reference and transconductance as the junction temperature reference, the two precursors are combined to obtain a temperature and package degradation independent estimate of the gate-oxide health. High-temperature gate bias (HTGB) and DC power cycling tests are used to confirm <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">g_m</i> ’s sensitivity to gate-oxide degradation and insensitivity to package degradation in kelvin-sourced discrete SiC MOSFETs. A simple circuit is proposed for on-board <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">g_m</i> measurement using the voltage drop on the common source inductance of the MOSFET. The measured <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">g_m</i> is converted into digital pulse width, which can be easily measured using a microcontroller. A comprehensive comparison is presented to demonstrate the merits of the proposed method. Lastly, the accuracy and sensitivity of temperature and aging measurements using the proposed method is validated experimentally.

Enhancing voice biometric security: Evaluating neural network and human capabilities in detecting cloned voices

This study assesses speaker verification efficacy in detecting cloned voices, particularly in safety-critical applications such as healthcare documentation and banking biometrics. It compares deeply trained neural networks like the Deep Speaker with human listeners in recognizing these cloned voices, underlining the severe implications of voice cloning in these sectors. Cloned voices in healthcare could endanger patient safety by altering medical records, causing inaccurate diagnoses and treatments. In banking, they threaten biometric security, increasing the risk of financial fraud and identity theft. The research reveals the neural network's superiority over human detection in pinpointing cloned voices, underscoring the urgent need for sophisticated AI-based security. The study stresses the importance of developing robust defenses against voice cloning attacks, which can have critical consequences in healthcare and fintech. This research is crucial for enhancing security in areas reliant on voice authentication, safeguarding confidential data, and preserving the integrity of vital services. The Polish National Center for Research and Development (NCBR) initially supported the project “BIOPUAP” (POIR.01.01.01-0092/19), which focused on digital banking. Subsequently, the project “ADMEDVOICE” (INFOSTRATEG4/0003/2022), also supported by the NCBR, conducted further research into voice cloning in the healthcare sector.

Diagnosis of Multicomponent Faults in SRM Drives Based on Auxiliary Current Reconstruction Under Soft-Switching Operation

Switched reluctance motors (SRMs) are considered a potential candidate for utilization in numerous safety-critical applications including electric vehicles (EVs) and more/all electric aircraft (MEA), etc., where the reliability of the drive system is critically important. An SRM drive system suffers from various electrical, mechanical, and sensor faults during operation. The existing fault detection schemes investigated each fragile component separately and proposed different approaches for each fault, which not only make the system complex but also less reliable for real-time implementation in safety-critical applications. Thus, a more sophisticated algorithm to detect and discriminate each of these faults in real time becomes attractive. A simple fault detection and isolation scheme for multiple faults in SRM drive systems is reported in this paper. The investigated faults include open- and short-circuit faults in power switches, open-phase winding, and zero-output current sensors faults. An auxiliary phase current for each phase was reconstructed from the dc-link and the faults are identified and discriminated effectively through the digital analysis of the fundamental current and the auxiliary current in real-time. The feasibility of the proposed scheme is verified by the extensive simulation and experiments conducted on a three-phase 12/8 SRM drive system.

Pore evolution mechanisms during directed energy deposition additive manufacturing

Porosity in directed energy deposition (DED) deteriorates mechanical performances of components, limiting safety-critical applications. However, how pores arise and evolve in DED remains unclear. Here, we reveal pore evolution mechanisms during DED using in situ X-ray imaging and multi-physics modelling. We quantify five mechanisms contributing to pore formation, migration, pushing, growth, removal and entrapment: (i) bubbles from gas atomised powder enter the melt pool, and then migrate circularly or laterally; (ii) small bubbles can escape from the pool surface, or coalesce into larger bubbles, or be entrapped by solidification fronts; (iii) larger coalesced bubbles can remain in the pool for long periods, pushed by the solid/liquid interface; (iv) Marangoni surface shear flow overcomes buoyancy, keeping larger bubbles from popping out; and (v) once large bubbles reach critical sizes they escape from the pool surface or are trapped in DED tracks. These mechanisms can guide the development of pore minimisation strategies.

Battery Remaining Useful Life Prediction Using Machine Learning Models: A Comparative Study

Predicting the remaining useful life (RUL) of lithium-ion (Li-ion) batteries is crucial to preventing system failures and enhancing operational performance. Knowing the RUL of a battery enables one to perform preventative maintenance or replace the battery before its useful life expires, which is vital in safety-critical applications. The prediction of the RUL of Li-ion batteries plays a critical role in their optimal utilization throughout their lifetime and supporting sustainable practices. This paper conducts a comparative analysis to assess the effectiveness of multiple machine learning (ML) models in predicting the capacity fade and RUL of Li-ion batteries. Three case studies are analyzed to assess the performances of the state-of-the-art ML models, considering two distinct datasets. These case studies are conducted under various operating conditions such as temperature, C-rate, state of charge (SOC), and depth of discharge (DOD) of the batteries in Cases 1 and 2, and a different set of features and charging policies for the second dataset in Case 3. Meanwhile, diverse extracted features from the initial cycles of the second dataset are considered in Case 3 to predict the RUL of Li-ion batteries in all cycles. In addition, a multi-feature multi-target (MFMT) feature mapping is introduced to investigate the performance of the developed ML models in predicting the battery capacity fade and RUL in the entire life cycle. Multiple ML models that are developed for the comparison analysis in the proposed methodology include Random Forest (RF), extreme gradient boosting (XGBoost), light gradient-boosting machine (LightGBM), multi-layer perceptron (MLP), long short-term memory (LSTM), and attention-LSTM. Furthermore, hyperparameter tuning is applied to improve the performance of the XGBoost and LightGBM models. The results demonstrate that the extreme gradient boosting with hyperparameter tuning (XGBoost-HT) model outperforms the other ML models in terms of the root-mean-squared error (RMSE) and mean absolute percentage error (MAPE) of the battery capacity fade and RUL for all cycles. The obtained RMSE and MAPE values for XGBoost-HT in terms of cycle life are 69 cycles and 6.5%, respectively, for the third case. In addition, the XGBoost-HT model handles the MFMT feature mapping within an acceptable range of RMSE and MAPE, compared to the rest of the developed ML models and similar benchmarks.

Advancing Digital Twin-Based Collision Avoidance: A Comprehensive Analysis of Communication Networks for Safety-Critical Applications in Industry 4.0.

This study presents a theoretical framework for defining the performance level of wireless safety functions within industrial environments. While acknowledging the simplifications inherent in our approach-primarily based on packet loss rates as a measure of system performance-the study underscores the dynamic challenges posed by real-world warehouses. Through an in situ measurement study of a forklift truck safety system, we validate the proposed method and emphasize the need for a more nuanced examination of wireless communication in complex settings. The study advocates for an expanded theoretical framework that considers fluctuations in warehouse dynamics, accounting for their impact on packet loss rates and, consequently, the precision of performance-level assessments. Furthermore, the research highlights the complexity introduced by wireless system characteristics not addressed in the simplified model, urging future investigations to incorporate these factors for a comprehensive understanding of wireless safety systems. The absence of specific criteria for wireless systems within existing standards emphasizes the necessity for a specialized framework in addressing safety aspects unique to wireless applications.

VFL-Chain: Bulletproofing Federated Learning in the V2X environments

Federated Learning (FL) has gained significant traction as a promising approach to enable collaborative machine learning (ML) while safeguarding data privacy across diverse applications, with the Vehicle-to-Everything (V2X) environment being a notable use case. However, conventional FL systems remain susceptible to model poisoning attacks, wherein malicious participants can introduce inaccurate or deliberately misleading local model updates to the central aggregator. In the context of V2X, such attacks could spawn catastrophic outcomes and jeopardize safety-critical applications. To tackle this pressing concern, we introduce a privacy-preserving and verifiable FL framework, dubbed VFL-Chain, designed to facilitate secure and efficient collaboration among intelligent connected vehicles (ICVs). VFL-Chain aspires to deliver a privacy-centric, verifiable FL experience complemented by a robust incentive mechanism tailored for ICVs. Our proposed solution leverages Bulletproofs to enable efficient verification of model update integrity, which are assimilated within smart contracts on an underlying permissioned blockchain. Furthermore, VFL-Chain presents a fair incentive mechanism that rewards honest participation, fostering a resilient and efficient FL implementation. We conduct an exhaustive security analysis and performance assessment of our proposed system, which underscores its efficacy in countering data poisoning attacks and augmenting the accuracy of FL.

Reliability of AI Algorithms in Safety Applications

The integration of Artificial Intelligence (AI) algorithms into safety-critical applications has become increasingly prevalent across various domains, including autonomous vehicles, medical diagnosis, industrial automation, and aerospace systems. These applications rely heavily on AI to make decisions that directly affect human safety, economic stability, and operational efficiency. Given the critical nature of these tasks, it is essential to rigorously assess the reliability of AI algorithms to ensure they perform consistently and accurately under all conditions. Reliability, in this context, refers to the AI system's ability to function without failure over a specific period, under defined operational conditions. In safety-critical domains, even minor errors or inconsistencies in AI decision-making can lead to catastrophic outcomes, such as traffic accidents involving autonomous vehicles, incorrect medical diagnoses leading to improper treatments, or failures in industrial processes that may cause costly downtime or even human casualties. The increasing complexity and deployment of AI technologies in these domains highlight the urgent need for a comprehensive understanding and evaluation of AI reliability. This paper provides a detailed analysis of the design considerations and methodologies for enhancing the reliability of AI algorithms. The discussion begins by exploring the underlying principles of reliability in AI systems, focusing on both theoretical and practical perspectives. We examine key factors that influence reliability, including data quality, algorithmic robustness, model interpretability, and system integration. The paper then delves into various reliability assessment techniques, such as fault tolerance mechanisms, error detection and correction methods, redundancy, and validation processes. To provide a deeper understanding of reliability in AI, we introduce mathematical models and statistical evaluation techniques that quantify reliability metrics. For instance, reliability modeling using exponential distribution, Monte Carlo simulations for probabilistic reliability analysis, and error propagation studies using Jacobian matrices are presented. We also explore the use of machine learning-specific reliability metrics, such as the Area Under the Curve (AUC) in Receiver Operating Characteristic (ROC) analysis, which helps evaluate the performance of AI in critical decision-making contexts. Furthermore, this paper addresses the current challenges and limitations in ensuring AI reliability, including computational complexity, ethical considerations, and regulatory compliance issues. We highlight the difficulties in developing AI models that can maintain their reliability across diverse and unpredictable real-world scenarios. The potential for bias, lack of transparency in AI decision-making, and difficulties in explaining complex AI models also present significant hurdles that need to be addressed to enhance reliability. The findings and methodologies discussed in this paper aim to contribute to a deeper understanding of the complex landscape of AI reliability, providing a framework for researchers, practitioners, and policymakers to develop safer, more reliable AI systems that can be trusted to operate in environments where safety is paramount.

Design and analysis of fault-tolerant sequential logic circuits for safety-critical applications

Safety-critical systems used in applications that demand high levels of dependability, efficiency, and fault-tolerance often use sequential logic circuits in its design and implementation. The safety-critical digital system typically uses latches, flip-flops, and other memory elements, which are prone to the effects of natural faults and single event upsets (SEUs) caused by radiation-induced effects. The faults can lead to subsystem failures due to the continuous advancement in the realization of the small size transistor. To design a reliable digital-based system, it is essential to develop new fault-tolerance approaches that are integrated into the design of sequential logic circuits. This work proposes a novel fault-tolerant approach based on the redundancy of sequential logic circuit, which consists of a variety of design components, D flip-flop storage elements linked to a fault injection unit, a duplicate modular redundancy, and data monitoring units with a switching circuit. The experimental simulation results using a five-state Markov chain analysis model prove that the proposed fault-tolerant system can achieve 0.99999998 for reliability of the fault detection coverage (C) which equal to 0.99999. Finally, we believe that using this new approach of fault-tolerance and redundancy would improve the dependability and reliability of next generation safety-critical applications.