1964–1970

Fault Tolerant System Safety

Contemporary themes

The period fused probabilistic reliability engineering with a systemic safety perspective, unifying Monte Carlo simulation, Markov models, and evaluation protocols across power, nuclear, and computer systems. Fault detection and redundancy emerged as core safety strategies, including sequential fault detection, fault-tolerant design, standby redundancy with repair, and fault analysis shaping safer engineering systems. Vehicle safety engineering concentrated on occupant protection through rear-end collision prevention, energy-absorption design, and restraint effectiveness, while hazard analysis under incomplete data and stochastic modelling broadened the reliability toolbox for partial information scenarios and power networks.

• System reliability engineering and probabilistic modelling unify across diverse engineering domains through mathematical methods, Monte Carlo simulation, Markov processes, and evaluation protocols, as shown in power, nuclear, and computer-system studies[6][8][10][11][15][18][20].
• Fault detection and redundancy emerge as core safety-enhancing strategies, with sequential-machine fault detection, fault-tolerant design, standby redundancy with repair, and fault analysis shaping safer engineering systems[5][12][14][16].
• Vehicle safety engineering concentrates on occupant protection through rear-end collision prevention, energy-absorption design, and restraint effectiveness evidenced in large accident analyses and protective components[2][4][9].
• Hazard analysis and incomplete data motivate robust reliability methods, including hazard plotting, discrimination between failure-density assumptions, and Monte Carlo-based confidence limits enabling safer predictions from partial information[7][11][17].
• Power transmission reliability evaluation leverages stochastic modelling (Markov methods, demand modelling) and evaluation protocols to assess network resilience and planning needs in electrical infrastructures[6][8][14][16][20].

Influential works

Hazard Plotting for Incomplete Failure Data introduced hazard plots to visualize risks with partial data, foreshadowing data-imputation approaches in reliability analysis and probabilistic safety. Unsafe at Any Speed exposed auto-design vulnerabilities and catalyzed regulatory reforms and a lasting safety culture in automotive design. The Costs of Accidents quantified the economic toll of injuries and property damage, legitimizing safety investments and risk-cost analysis in policy and business.

• Unsafe at Any Speed (1966)[3] exposed auto-design vulnerabilities and systemic safety risks; it catalyzed consumer advocacy, regulatory reforms, and a safety-driven industry culture that persists in automotive standards today.
• The Costs of Accidents (1970)[4] quantified the economic toll of injuries and property damage, legitimizing safety investments and risk-cost analysis in policy and business, shaping later safety economics methods.
• Hazard Plotting for Incomplete Failure Data (1969)[6] proposed hazard plots to visualize risks with partial data, pioneering data-imputation-leaning reliability analysis and influencing later probabilistic safety approaches.
• Fault detecting experiments for sequential circuits (1964)[8] established early methods for diagnosing faults in sequential hardware, underpinning reliability engineering and safe design of digital systems.
• A Method for the Design of Fault Detection Experiments (1970)[12] provided a structured framework for planning fault-detection tests, enabling targeted verification of safety-critical systems and advancing hardware reliability research.

1971–1977

Modular Dependable Systems

Contemporary themes

During 1971–1977, the field converged on designing systems that remain functional under faults through redundancy, repair strategies, and modular architectures. Researchers emphasized integrating hardware and software perspectives to enhance availability and safety, advancing practical redundancy in safety-critical systems and formalizing modular boundaries, distributed subsystems, and trusted cooperation to reduce coupling and support evolution. Methodologies for diagnosis and verification matured, with systematic fault detection, diagnosability analysis, and fault handling for intermittent faults in digital and software-driven platforms. Reliability analysis integrated fault trees with design optimization to guide safer trade-offs, while maintenance planning framed preventive strategies as optimization problems to maximize uptime under redundancy constraints.

• Fault tolerance is achieved through redundancy, standby architectures, and repair strategies, integrating hardware and software perspectives. Across theoretical fault-tolerant design (The STAR Computer) and practical redundancy in safety-critical systems, replication and repair scheduling improve availability and reliability[4][8][9][13][14].
• Prominent work formalizes modular decomposition and dependable architecture as core design principles for complex systems. Criteria for module boundaries, distributed subsystems, and trusted cooperation reduce coupling and support evolution in safety-critical platforms[3][4][16][20].
• Diagnosis and verification are addressed via systematic fault detection, diagnosis of diagnosable systems, and intermittent fault handling. The literature develops models and procedures for identifying faults in digital circuits and software-driven systems, emphasizing detection, isolation, and validation[1][2][7][10][17].
• Reliability analysis combines fault trees, risk assessment, and design optimization to improve safety and performance in complex systems. Foundational methods (fault tree analysis) are extended to model failures and guide optimal design choices for safer, more dependable systems[5][14][18].
• Maintenance planning is framed as an optimization problem to maximize system availability and mean time to first failure under redundancy constraints. Studies derive preventive maintenance policies, balancing costs, risk, and reliability for two-unit standby systems[9][11][12].

Influential works

• An innovations approach to fault detection and diagnosis in dynamic systems introduced an innovations-based framework for detecting faults in real-time by analyzing residuals between observed and expected behavior, enabling rapid diagnosis and safer operation[8]
• NEAR-MISS DETERMINATION THROUGH USE OF A SCALE OF DANGER proposed a structured near-miss classification scale to quantify hazards and trigger safety improvements before incidents, influencing risk assessment practices in systems safety[10]
• A LOGICAL FRAMEWORK FOR CATEGORIZING HIGHWAY SAFETY PHENomena AND ACTIVITY provides a formal taxonomy for highway safety phenomena and activities, enabling consistent data collection, analysis, and policy-making and serving as a cornerstone of highway safety engineering[3]
• Of Acceptable Risk: Science and the Determination of Safety argues for an empirical, scientifically grounded approach to determining acceptable risk, shaping regulatory standards and risk-based decision-making across engineering safety domains[2]
• Structured Systems Analysis: Tools and Techniques introduces Structured Systems Analysis as a formal methodology to decompose complex safety-critical systems, enabling systematic hazard analysis, risk reduction, and safer system design[7]

1978–2003

Robust Model-Based Diagnosis

Contemporary themes

During 1978–2003, robust model-based fault diagnosis unified analytical redundancy, parity relations, and observer-based detection to diagnose and isolate faults in dynamic systems, creating a cohesive theoretical framework. Uncertainty-aware failure detection and data-driven online methods, including wavelet-based analysis and real-time monitoring, advanced robust thresholds and rapid fault identification across mechanical and electrical domains. Structured safety analysis and design integration, encompassing expert systems and parity-based frameworks, provided a roadmap for safety-focused online diagnosis in critical systems, with domain-focused fault detection in induction motor drives illustrating practical applicability.

• Model-based fault diagnosis unifies analytical redundancy, parity equations, and observer-based detection to diagnose and isolate faults in dynamic systems, creating a cohesive theoretical framework[1][2][9][11][12][15].
• Uncertainty-aware failure detection analyzes model uncertainty and measurement variability to design robust thresholds and redundancy, improving reliability under uncertainty[3][4][5][20].
• Signal processing and data-driven online methods, including wavelet-based analysis and online motor fault monitoring, enable fast, real-time fault detection across machinery[14][16][17][18].
• Structured safety analysis and design integration, including expert systems and parity-based frameworks, provide a roadmap for safety-focused analysis and on-line diagnosis in critical systems[7][10][13].
• Domain-focused fault detection in induction motor drives and sensor systems illustrates applied methods across mechanical and electrical domains, linking theory to practice[6][8][14].

Influential works

The period solidified model-based residual generation as a core safety technology, laying groundwork for future diagnostic standards and fault-tolerant design. The emergence of safety lifecycle frameworks and probabilistic assessments reframed safety as a design- and management-level concern rather than solely a technical challenge. Pioneering developments in robust model-based fault diagnosis and probabilistic risk assessment influenced subsequent safety engineering practice and standardization.

• IEC 61508: functional safety of electrical/electronic/ programmable electronic safety-related systems: overview (1999)[3] established a comprehensive lifecycle framework and Safety Integrity Level (SIL) scheme for safety-related systems, standardizing risk assessment, design, verification and maintenance across industries; this standard profoundly shaped subsequent sector standards and practices.
• The Limits of Safety (1993)[11] argues that perfect safety is unattainable and that safety gains have diminishing returns, reframing safety as a management and design challenge; it influenced risk assessment methods and resilient, fault-tolerant thinking in engineering practice.
• The Concept of Probability in Safety Assessments of Technological Systems (1990)[10] provides a formal probabilistic framework for safety analysis, introducing methods to quantify failure likelihoods and system risk; it shaped probabilistic risk assessment and decision making in safety engineering.
• Robust Model-Based Fault Diagnosis for Dynamic Systems (1999)[4] presents robust fault diagnosis using model-based residual generation and observers, enabling early fault detection in uncertain dynamic systems and laying groundwork for safety-critical fault management.
• What Practices Will Most Improve Safety? (2002)[15] analyzes high-leverage management, organizational, and engineering practices that yield safety gains, informing industry safety programs, policy, and research on safety performance.

2004–2010

Systems-Theoretic Safety Engineering

Contemporary themes

Across 2004-2010, research converged on cross-domain system safety modeling that unifies accident causation, safety management, and safety climate across construction, logistics, software, and industrial engineering. Fault detection and diagnosis in rotating machinery and gearboxes leveraged time-frequency analysis, wavelets, Hilbert transforms, demodulation, and machine learning for online, rapid diagnostics in electrical and mechanical domains. Software and processor safety trends emphasized control-flow integrity, secure execution, and runtime protection, while empirical safety performance work quantified safety maturity and risk exposure in operational environments.

• Cross-domain system safety modeling combines accident models, safety management systems, and safety climate to unify practices across construction, logistics, software, and industrial engineering. These works trace a theoretical lineage from accident causation to organizational safety metrics[1][4][6][10][12].
• Fault detection and diagnosis in rotating machinery and gearboxes leverage time-frequency analysis, wavelets, Hilbert transforms, demodulation, and machine learning to enable on-line and rapid fault diagnosis across electrical and mechanical domains[2][3][5][8][9][13][14][16][19][20].
• Software and processor safety trends emphasize control-flow integrity, secure program execution, and fault-location techniques in large software systems, highlighting programming discipline and runtime protection as core safety mechanisms[7][15][17].
• Empirical safety performance work in construction and terminal operations uses factor-analysis, safety climate, and effectiveness measurements to quantify safety maturity and risk exposure in operational environments[1][4][6][10].

Influential works

Foundational contributions include Systems-Theoretic Accident Model and Process, which reframed accidents as outcomes of socio-technical interactions and reshaped hazard analysis. The 2008 survey Model-based Fault Diagnosis Techniques consolidated approaches for timely diagnosis in complex systems, accelerating adoption of model-based safety engineering. The human factor: the critical importance of effective teamwork and communication in providing safe care, Planetary Boundaries (2009), and Experimental Security Analysis of a Modern Automobile (2010) collectively extended safety thinking into governance, cyber-physical security, and teamwork, shaping long-term risk management.

• 'A new accident model for engineering safer systems' (2004)[3] introduces Systems-Theoretic Accident Model and Process (STAMP), a holistic view that treats accidents as outcomes of complex socio-technical interactions rather than simple sequences; it reshapes hazard analysis and safety design for modern systems.
• 'Model-based Fault Diagnosis Techniques: Design Schemes, Algorithms, and Tools' (2008)[6] surveys and consolidates model-based diagnosis methods, architectures, and tools, enabling timely fault detection, isolation, and diagnosis in complex systems; it elevated model-based safety engineering and influenced many subsequent diagnostic research.
• 'Planetary Boundaries: Exploring the Safe Operating Space for Humanity' (2009)[4] delineates quantifiable environmental thresholds that define a safe operating space; it introduced a cross-disciplinary safety framework linking earth-system science with governance, shaping long-term risk management and sustainability thinking.
• 'Experimental Security Analysis of a Modern Automobile' (2010)[11] demonstrates practical cyber-physical vulnerabilities in contemporary vehicles, showing attackers can manipulate safety-critical controls; it sparked renewed focus on automotive security, secure-by-design controls, and integrity monitoring.
• 'The human factor: the critical importance of effective teamwork and communication in providing safe care' (2004)[13] analyzes how teamwork, communication, and non-technical factors influence patient safety; it provided evidence for safety culture and informed system designs that reduce human error.

2011–2017

Safety Systems Resilience Engineering

Contemporary themes

During the 2011-2017 period, research in system safety fused data-driven condition monitoring and fault diagnosis for rotating machinery with advances in signal processing and machine learning. Techniques such as kurtogram, wavelet analysis, Ensemble Empirical Mode Decomposition, and deep learning were applied to motors, bearings, and gearboxes, complemented by real-time estimators and observer-based schemes enabling immediate fault detection in wind turbines and motor drives. The era also saw the emergence of robust multi-fault identification toolkits and KPI-driven decision support, unifying reliability engineering with holistic risk management and governance of complex systems.

• Data-driven condition monitoring and fault diagnosis of rotating machinery using signal processing and machine learning, spanning kurtogram, wavelet analysis, Ensemble Empirical Mode Decomposition (EEMD), and deep learning across motors, bearings and gearboxes[5][6][7][8][13][15][16][17].
• Real-time/open-circuit fault diagnosis in wind turbines and motor drives using online estimators and observer-based schemes, enabling immediate fault detection and mitigation[4][10][19].
• Advanced signal processing toolkits for robust multi-fault identification in rotating machinery, integrating kurtogram, high-resolution spectral analysis, EEMD, multiwavelet packets and stochastic resonance for feature extraction and fault estimation[2][6][7][8][11][13][14].
• System safety thinking and risk management framing safety across organizations and infrastructure, from organizational accident risk to metro construction warning systems, emphasizing holistic safety science and governance[1][3][12].
• Performance-oriented fault diagnosis using Key Performance Indicators (KPIs) and machine learning including Support Vector Machines (SVM) and pattern recognition to improve reliability and decision support in bearings, motors and drives[8][15][16][17][20].

Influential works

This era catalyzed a shift from conventional safety approaches toward resilience engineering, emphasizing work-as-done and performance variability as central to safety. Foundational cross-domain works framed safety as a socio-technical, systems-thinking challenge, influencing risk management, audits, and policy across organizations and infrastructure. The period laid essential groundwork for predictive maintenance and safety governance by integrating data-driven diagnostics with safety culture and governance concepts.

• Engineering a safer world: systems thinking applied to safety (2012)[2] introduced a holistic, socio-technical, systems-thinking framework for safety, linking design, operations, and risk governance; foundational for resilience and systemic risk management in safety engineering and shaping later safety culture and governance discourse.
• Safety-I and Safety-II: The Past and Future of Safety Management (2014)[4] reframes safety as performance variability, distinguishing failure prevention from successful work adaptation; catalyzed resilience engineering and broad systemic thinking in research and practice.
• From Safety-I to Safety-II: A White Paper (2014)[13] outlines a transition to Safety-II, emphasizing work-as-done, everyday adaptations, and learning diagnostics; widely cited as a practical pathway for organizations and regulators.
• Managing the Risks of Organizational Accidents (2016)[1] extends the Swiss cheese model to organizational failures, highlighting culture, reporting, and systemic defenses; its enduring influence on risk management, audit, and safety policy.
• Real-Time Motor Fault Detection by 1-D Convolutional Neural Networks (2016)[11] applies Convolutional Neural Networks (CNNs) to vibration signals for rapid fault diagnosis in motors, enabling predictive maintenance and safer operation of safety-critical systems.

2018–2024

Domain-Adaptive Resilient System Safety

Contemporary themes

The period from 2018 to 2024 saw fault diagnosis and safety analytics pivot toward data-driven learning, with convolutional neural networks, autoencoders, and recurrent networks extracting discriminative features from vibration, acoustic, and image signals across bearings, gears, and motors. Researchers addressed noise, imbalanced classes, and load variability, while domain adaptation and transfer learning bridged distribution shifts across machines and environments, enabling knowledge transfer without full retraining. Adversarial learning, ensemble generalization, and robust feature integration with signal-processing techniques enhanced reliability under changing operating conditions; data-efficient strategies and hybrid architectures became common to address scarce labels and multimodal inputs.

• Data-driven deep learning for fault diagnosis leveraging CNNs, autoencoders, and RNNs to learn discriminative representations from vibration, acoustic, or image signals across bearings, gearboxes, and motors, addressing noise, imbalanced classes, and load variability[1][2][3][4][5][10][12][13][15][16][17].
• Cross-domain fault diagnosis via domain adaptation and transfer learning to bridge distribution shifts arising from different machines, loads, and operating environments; enabling knowledge transfer without retraining from scratch[6][14].
• Adversarial learning and ensemble-generalization approaches to enhance robustness of fault diagnosis models under varied operating conditions, including adversarially trained CNNs and capsule-net based generalization[8][12].
• Integration of signal-processing-based feature extraction (multi-scale morphology, impulse wavelets, entropy measures) with ML classifiers to improve fault detection and interpretability in rotating machinery[11][18][20].
• Hybrid architectures and data-efficient strategies combining deep learning with traditional ML ensembles, compressed data acquisition, or multimodal inputs to boost reliability and performance under scarce labeled data[4][16].

Influential works

• Safety–I and Safety–II (2018) reframes safety from a failure-centric view to resilience engineering, introducing Safety-I (avoiding losses) and Safety-II (ensuring successful operation) and stressing performance variability and human factors; it reshaped risk governance across aviation, healthcare, and industry[8].
• Toward safer highways demonstrates the integration of Extreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP) to detect real-time traffic accidents and reveal key drivers; this work advanced interpretable ML for roadway safety and influenced subsequent ITS analytics[12].
• Flights to Safety analyzes aviation safety data to identify systemic risk patterns, margins, and mitigations; it introduces risk-based metrics and safety envelopes that influenced policy and data-driven improvements in air travel[1].
• Deep Residual Shrinkage Networks (DRSN) propose noise-robust residual architectures that adaptively suppress outliers, yielding superior fault diagnosis in noisy industrial settings and inspiring later DL-based safety diagnostics in engineering systems[5].
• A Generic Intelligent Bearing Fault Diagnosis System Using Compact Adaptive 1D CNN Classifier shows that One-Dimensional Convolutional Neural Networks (1D-CNNs) can provide compact, fast bearing fault diagnosis, enabling effective condition monitoring and informing predictive maintenance in safety-critical equipment[14].