Joel B. Harley – SmartDATA Lab

When AI Says “Maybe”: The Quest for Meaningful Uncertainty in Machine Learning

In the realm of artificial intelligence (AI), particularly within machine learning (ML), the ability to quantify uncertainty is paramount. As AI systems increasingly influence critical decisions in fields like healthcare, engineering, and finance, understanding the confidence of these systems becomes essential. Yet, translating the abstract probabilities of AI models into actionable insights remains a significant challenge.

Joel B. Harley June 25, 2025 0

Statistical Partial Wavefield Imaging Repository

Statistical Partial Wavefield Imaging (SPWI) is a powerful signal processing algorithm designed to detect and localize structural damage using sparse ultrasonic sensor arrays. Unlike traditional imaging methods that require dense sensor grids or full wavefield measurements, SPWI extracts high-quality damage localization images from partial and limited measurements, making it ideal for real-world Structural Health Monitoring (SHM) applications.

Joel B. Harley June 24, 2025 0

The Next Frontier in AI: Merging Physics with Data for Smarter Models

In the realm of artificial intelligence (AI), a new paradigm is emerging—one that marries the empirical rigor of physics with the adaptability of data-driven models. This hybrid approach is poised to revolutionize industries and scientific research by addressing the limitations inherent in purely data-driven or purely physics-based models.

Joel B. Harley June 18, 2025 0

Dynamic Time Warping for Temperature Compensation Repository

Guided wave Structural Health Monitoring (SHM) systems are powerful tools for detecting damage in structures like aircraft fuselages, pipelines, and bridges. However, one major obstacle persists: temperature variation. Even small temperature changes can significantly distort guided wave signals, often leading to false positives or missed damage.

Enter Dynamic Time Warping (DTW) Temperature Compensation — a robust, data-driven algorithm developed by Alexander Douglass and Dr. Joel B. Harley that aligns guided wave signals distorted by temperature, improving the accuracy and reliability of damage detection without requiring physical models or manual calibration.

Joel B. Harley June 17, 2025 0

Why AI Can’t Yet Grow a Perfect Crop Model

Imagine trying to beat Elden Ring with only half the map, no health potions, and a sword that breaks every few swings. That’s roughly the challenge agricultural scientists face when applying artificial intelligence (AI) to crop modeling.

Crop models—like the venerable Decision Support System for Agrotechnology Transfer (DSSAT)—simulate plant growth, soil chemistry, and environmental interactions. They’re essential tools for predicting yields, managing fertilizers, and preparing for climate change. But these models are only as good as their inputs, and in agriculture, data is often scarce, noisy, or incomplete.

Joel B. Harley June 11, 2025 0

K-SVD Dictionary Learning for Damage Detection Repository

K-SVD Dictionary Learning for Damage Detection is a baseline-free, data-driven approach for detecting structural damage using guided ultrasonic waves. Developed by Supreet Alguri and Dr. Joel B. Harley, this method eliminates the need for pristine baseline measurements—a major limitation in many real-world Structural Health Monitoring (SHM) systems.

This CodeOcean capsule provides a reproducible implementation of the algorithm, as described in the paper:

Joel B. Harley June 10, 2025 0

Why AI Struggles to Crack the Code of Nondestructive Testing

There’s a common trope in science fiction: an all-seeing machine, unblinking and exact, scanning the world for flaws we can’t detect. In the real world of nondestructive testing (NDT)—the science of using sensors to find cracks, corrosion, and hidden damage in everything from aircraft wings to oil pipelines—that dream still feels frustratingly out of reach.

We have artificial intelligence that can beat world champions at Go, generate Hollywood-quality dialogue, and diagnose rare diseases from pixelated scans. And yet, when we point these same tools at ultrasonic signals or thermographic images from NDT inspections, the results are… inconsistent at best.

Joel B. Harley June 4, 2025 0

Temporal Sparse Wavenumber Analysis Repository

Temporal Sparse Wavenumber Analysis (TSWA) is a novel technique that reconstructs high-resolution spatiotemporal wavefields using fewer temporal samples than traditional methods typically require. Developed by Soroosh Sabeti and Dr. Joel B. Harley, TSWA makes it possible to retrieve accurate guided wave information from temporally undersampled data, which is crucial in applications where high-speed sensing, data storage, or power consumption are limiting factors.

Joel B. Harley June 3, 2025 0

When Biomechanics Meets Reality: The OpenSim Simulation Gap

In the realm of biomechanics, OpenSim stands as a cornerstone—a powerful, open-source simulation engine that enables researchers to model and analyze musculoskeletal dynamics. From studying gait patterns to designing prosthetics, OpenSim has been instrumental in advancing our understanding of human movement.

Yet, as with any model, there exists a gap between simulation and reality. Recent studies have highlighted discrepancies between OpenSim’s predictions and actual sensor data, raising questions about the fidelity of these simulations. This divergence, often referred to as the “sim-to-real gap,” underscores the challenges inherent in accurately modeling complex biological systems.

Joel B. Harley May 21, 2025 0

Data-Driven Matched Field Processing (DDMFP) Repository

Data-Driven Matched Field Processing (DDMFP) is an innovative signal processing framework designed for localizing acoustic sources in complex environments, such as those encountered in structural health monitoring (SHM) using Lamb waves. Traditional matched field processing (MFP) techniques rely heavily on accurate physical models of the propagation medium, which can be challenging to obtain in real-world scenarios. DDMFP circumvents this limitation by constructing localization models directly from measured data, enhancing robustness and accuracy in complex, multimodal propagation environments.

Joel B. Harley May 17, 2025 0

Author: Joel B. Harley