Chandrashekhar Nataraj

Objectives: Evaluate the association between cardiopulmonary resuscitation (CPR) ventilation rates and survival to hospital discharge, and identify novel ventilation targets. Design: Multicenter prospective ancillary observational cohort study of the ICU-RESUS trial (NCT02837497). Setting: A total of 18 ICUs from 10 U.S. clinical sites from October 2016 to March 2021. Patients: Hospitalized children (≤ 18 yr) with endotracheal or tracheostomy tubes and evaluable intra-arrest end-tidal carbon dioxide data to calculate ventilation rate. Interventions: None. Measurements and Main Results: Among 234 included events, 86 events (36.8%) had guideline-recommended (20–30 beats per min [bpm]) average ventilation rates during CPR. The association between guideline-recommended CPR ventilation rate and survival to hospital discharge was evaluated with Poisson regression using generalized estimating equations, controlling for a priori covariables (initial pulse status, immediate cause of arrest). We found no association between guideline-recommended ventilation rates and survival to hospital discharge (adjusted relative risk [aRR], 0.95; 95% CI, 0.75–1.21; p = 0.68). In exploratory analyses, novel age-stratified ventilation rates were identified using natural cubic splines and subsequently evaluated in multivariable models. We identified potential thresholds (< 8 yr, ≥ 26 bpm; ≥ 8 yr, < 26 bpm). For children younger than 8 years, average CPR ventilation rate greater than or equal to 26 bpm was associated with improved survival to hospital discharge compared with less than 26 bpm (aRR, 1.32; 95% CI, 1.00–1.73; p = 0.048). Conclusions: CPR ventilation rates in the currently recommended range (20–30 bpm) were not associated with survival in children with in-hospital cardiac arrest. In exploratory analyses in children younger than 8 years, higher ventilation rates (≥ 26 bpm) were associated with improved survival and warrant confirmation in future studies.

Artificial intelligence (AI) is rapidly transforming mechanical engineering (ME) practice, influencing activities that range from modeling and simulation to optimization, diagnostics, design, and technical communication. This article consolidates perspectives discussed at a dedicated round-table during the 28th ABCM International Congress on Mechanical Engineering—COBEM 2025—and offers a structured reflection for the Brazilian and international ME communities. Rather than providing an exhaustive survey, the paper develops a coherent narrative around the conditions under which AI can strengthen—rather than dilute—engineering rigor. The article begins by clarifying how contemporary AI systems operate, distinguishing statistical learning from human cognition and outlining their capabilities, limitations, and emerging agentic AI systems. It then examines industrial adoption, workforce transformation, and the growing role of AI in scientific discovery and engineering workflows. Particular attention is devoted to the energy and infrastructure demands of AI-driven computation and their implications for sustainability, power systems, and thermal management. Building on these constraints, the paper discusses integrating physical knowledge, structured reasoning, and domain expertise into AI systems for safety-relevant engineering applications. Emphasis is placed on representative ME applications, physics-informed and hybrid learning frameworks, model development under physical constraints, validation, uncertainty quantification, and certification challenges. The paper also addresses the educational, epistemological, and ethical implications of AI, including curriculum modernization, human-agent collaboration, professional accountability, and the long-term sustainability of engineering knowledge. It concludes with a synthesis of principles for the responsible adoption of AI in ME, encompassing physical consistency, validation, awareness of uncertainty, energy sustainability, human oversight, and ethical responsibility. Our central message is that AI should be viewed not as a substitute for engineering judgment, but as a powerful tool whose value depends on disciplined integration with physical reasoning, scientific rigor, and professional responsibility.

Continuous monitoring of industrial rotating machines is essential for ensuring health, reliability, and safety, while minimizing costs from unplanned downtime. Traditional systems require domain experts to configure parameters for each machine, limiting adaptability. Machine learning techniques offer alternatives, with supervised approaches demanding labeled datasets that are rarely available due to the extensive effort required from specialists. This study presents an algorithm for unsupervised condition monitoring based on an online distribution-driven clustering approach. The algorithm autonomously learns machine behavior, detecting new operating conditions in real time while recognizing previously observed states. Input data consist of signals from proximity and accelerometer sensors mounted on the rotor shaft. These signals are processed using time- and frequency-domain feature extraction methods combined with an autoencoder neural network. Validation was performed with four datasets: synthetically generated signals, real measurements from a rotating machine at the Federal University of Uberlandia, and benchmark datasets from Case Western Reserve University (CWRU) and Hanoi University of Science and Technology (HUST) bearing dataset. Evaluation metrics, including accuracy and F1 score, consistently reached values close to 0.96 across most tests. These results confirm the algorithm’s ability to reliably detect and classify machine condition variations in real time. Furthermore, results obtained with benchmark datasets were compared to supervised learning studies, showing comparable or superior performance. This demonstrates the proposed method’s robustness and practicality in industrial applications, where labeled data are scarce and real-time adaptability is critical. Overall, the algorithm provides an effective solution for unsupervised monitoring of rotating machinery, reducing reliance on expert intervention while maintaining high detection accuracy.

Medical diagnostics continues to be one of the most difficult challenges in healthcare, with diagnostic errors constituting the most common, costly, and harmful category of medical errors. They contribute to millions of adverse outcomes globally each year. The principal difficulty lies in the extraordinary complexity of the human body, a multiscale, adaptive, nonlinear dynamical system whose hidden states defy simplifications and contradict intuitive thinking. Current practice, largely dependent on heuristic guidelines, physician judgment, and black box machine learning, remains fundamentally limited, perpetuating diagnostic failures and preventing true personalization. This paper argues that nonlinear mechanics and dynamics are not just refinements but essential to understanding physiology. Nonlinear phenomena such as instabilities, bifurcations, chaos, fractals, adaptive feedback, and multiscale interactions occur across all the systems in the body including cardiovascular, respiratory, metabolic, neural, immune, and musculoskeletal subsystems, and are central to both health and disease. Ignoring these phenomena costs us mechanistic understanding and puts accurate diagnostics out of reach. At the same time, mechanistic models, data-driven Artificial Intelligence, and physician expertise each have unique strengths but are inadequate when applied in isolation. We propose their synthesis through physics-informed machine learning, hybrid frameworks, and the emerging paradigm of digital twins. Such systems combine mechanistic insights, data-driven computations, and experiential clinical wisdom to deliver interpretable and personalized diagnostics. Importantly, embedding nonlinear mechanics in real-time, patient-specific, hybrid models provides an exciting path toward reducing errors, improving outcomes, and transitioning from reactive, guideline-driven practice to truly pro-active, precision medicine.

Traumatic brain injury (TBI) presents a significant public health challenge, often resulting in mortality or lasting disability. Predicting outcomes such as mortality and Functional Status Scale (FSS) scores can enhance treatment strategies and inform clinical decision-making. This study applies supervised machine learning (ML) methods to predict mortality and FSS scores using a real-world dataset of 300 pediatric TBI patients from the University of Colorado School of Medicine. The dataset captures clinical features, including demographics, injury mechanisms, and hospitalization outcomes. Eighteen ML models were evaluated for mortality prediction, and thirteen models were assessed for FSS score prediction. Performance was measured using accuracy, ROC AUC, F1-score, and mean squared error. Logistic regression and Extra Trees models achieved high precision in mortality prediction, while linear regression demonstrated the best FSS score prediction. Feature selection reduced 103 clinical variables to the most relevant, enhancing model efficiency and interpretability. This research highlights the role of ML models in identifying high-risk patients and supporting personalized interventions, demonstrating the potential of data-driven analytics to improve TBI care and integrate into clinical workflows.