Abstract The increasing complexity of automotive Electronic Control Units (ECUs) demands integrated design approaches that simultaneously address electrical performance, thermal reliability, functional safety, and real time validation. However, the scarcity of public multi-physics datasets has hindered the development of AI-driven optimization frameworks capable of unifying these competing objectives. This study presents the development and comprehensive analysis of a synthetic power thermal dataset for AI-driven ECU... The increasing complexity of automotive Electronic Control Units (ECUs) demands integrated design approaches that simultaneously address electrical performance, thermal reliability, functional safety, and real time validation. However, the scarcity of public multi-physics datasets has hindered the development of AI-driven optimization frameworks capable of unifying these competing objectives. This study presents the development and comprehensive analysis of a synthetic power thermal dataset for AI-driven ECU design optimization and risk classification. The dataset, generated through physics-informed simulations, comprises over 15 electrical, thermal, and composite variables representing realistic ECU operating conditions. Through extensive statistical testing, regression modeling, and correlation analysis including Pearson and Spearman correlations, linear and polynomial regression, and multi-collinearity assessment we identify the dominant physical drivers of thermal stress and risk behavior. Results reveal that while voltage and electrical resistance exhibit negligible influence on current variation and IR drop (R² = 0.000461, p = 0.0205), Power Density, Current, and total Power are the paramount factors dictating temperature rise and thermal stress, with Current exhibiting a very strong correlation with Thermal Stress (r = 0.810). The proposed closed loop Hardware in the Loop (HIL) integrated AI pipeline incorporating multi-task learning, FPGA accelerated inference, and physics informed neural networks enables simultaneous multi class risk classification and predictive thermal modeling. This work provides a foundational benchmark for next generation ECU design, indicating that physics aware synthetic datasets, combined with hybrid AI architectures, can effectively support real time, safety compliant optimization while reducing development costs and physical prototyping requirements. Read More Read Less

Abstract Consumer grade Micro Electro Mechanical Systems (MEMS) accelerometers are ubiquitous in autonomous navigation platforms but remain susceptible to inherent stochastic noise and deterministic biases. This study presents a comprehensive empirical noise profiling and dimensionality reduction analysis of raw, high-frequency triaxial acceleration data acquired from an MPU-6050 sensor under controlled, static conditions. A continuous dataset comprising 37,459 samples was captured using a Raspberry Pi Pico microcontroller.... Consumer grade Micro Electro Mechanical Systems (MEMS) accelerometers are ubiquitous in autonomous navigation platforms but remain susceptible to inherent stochastic noise and deterministic biases. This study presents a comprehensive empirical noise profiling and dimensionality reduction analysis of raw, high-frequency triaxial acceleration data acquired from an MPU-6050 sensor under controlled, static conditions. A continuous dataset comprising 37,459 samples was captured using a Raspberry Pi Pico microcontroller. The analytical framework employed descriptive statistics, dual method anomaly detection (Interquartile Range and Z-score), and Principal Component Analysis (PCA) to rigorously evaluate signal integrity and multivariate covariance structures. Statistical profiling revealed a significant factory bias offset along the X axis (approximately 1,432 LSB) and an elevated, independent noise floor along the Z-axis (standard deviation of 493.2 LSB), in contrast to the stable, gravity aligned Y-axis. Furthermore, anomaly detection highlighted transient data corruption events, likely stemming from I²C communication glitches or power fluctuations. Crucially, PCA demonstrated that 75.1% of the total dataset variance is captured by the first two components: PC1 captures shared systemic noise across the X and Y axes and magnitude error, while PC2 isolates the distinct, orthogonal dynamics of the Z-axis. These findings validate the efficacy of PCA-based dimensionality reduction and targeted, axis specific filtering. By projecting high dimensional sensor streams onto principal components, resource constrained embedded systems can achieve substantial computational savings while preserving essential signal characteristics, ultimately facilitating the development of resilient, self calibrating inertial navigation algorithms. Read More Read Less

Abstract Accurate short-term load forecasting is essential for the efficient and reliable operation of modern smart grids. This study develops and comprehensively evaluates a high-resolution, interpretable load forecasting framework using 10-minute smart meter data from Laâyoune Zone 1, Morocco. To capture complex temporal dependencies, we engineered advanced features, including lag variables, rolling statistics, and cyclical time encodings. Multiple machine learning models were systematically compared, with... Accurate short-term load forecasting is essential for the efficient and reliable operation of modern smart grids. This study develops and comprehensively evaluates a high-resolution, interpretable load forecasting framework using 10-minute smart meter data from Laâyoune Zone 1, Morocco. To capture complex temporal dependencies, we engineered advanced features, including lag variables, rolling statistics, and cyclical time encodings. Multiple machine learning models were systematically compared, with the Random Forest (RF) ensemble achieving the best overall performance (RMSE: 5.42, R²: 0.87), effectively capturing nonlinear demand dynamics. Notably, the unexpectedly strong performance of the Persistence baseline highlighted the significant short-term autocorrelation inherent in high-frequency residential and industrial load data. Multi-step forecast horizon analysis revealed exceptional accuracy for 10 to 60 minute predictions, making the models highly suitable for real-time grid balancing and transformer management, although uncertainty gradually accumulates beyond two hours. Residual diagnostics further confirmed minimal systematic bias and stable model behavior across varying demand conditions. Furthermore, SHAP-based interpretability analysis confirmed that recent lagged load measurements and temporal indicators are the primary drivers of predictions, overcoming the traditional “black-box” limitations of machine learning. Ultimately, this explainable ensemble framework provides a robust, transparent, and computationally efficient tool for utility operators, facilitating demand response, anomaly detection, and sustainable energy management in modern smart grid environments. Read More Read Less