Machine-learning (ML) techniques hold the potential of enabling efficient quantitative micrograph analysis, but the robustness of ML models with respect to real-world micrograph quality variations has not been carefully evaluated. We collected thousands of scanning electron microscopy (SEM) micrographs for molecular solid materials, in which image pixel intensities vary due to both the microstructure content and microscope instrument conditions. We then built ML models to predict the ultimate compressive strength (UCS) of consolidated molecular solids, by encoding micrographs with different image feature descriptors and training a random forest regressor, and by training an end-to-end deep-learning (DL) model. Results show that instrument-induced pixel intensity signals can affect ML model predictions in a consistently negative way. As a remedy, we explored intensity normalization techniques. It is seen that intensity normalization helps to improve micrograph data quality and ML model robustness, but microscope-induced intensity variations can be difficult to eliminate.

Monitoring mechanical systems operating in uncertain environments contaminated with both environmental disturbances and noise lead directly to low signal-to-noise-ratios, creating an extremely challenging processing problem, especially in real-time. In order to estimate the performance of a particular system from uncertain vibrational data, it is necessary to identify its unique resonant (modal) frequency signature. The monitoring of structural modes to determine the condition of a device under investigation is essential, especially if it is a critical entity of an operational system. The development of a model-based scheme capable of the on-line tracking of the inherent structural modal frequencies by applying both constrained subspace identification techniques to extract the modal frequencies and state estimation methods to track the evolution is discussed. An application of this approach to a cylindrical structural device (pipe-in-air) is analyzed based on theoretical simulations along with controlled validation experiments, including injected anomalies illustrate the approach and performance. Statistics are gathered to bound potential processors for real-time performance employing these constrained techniques.

Dynamic testing of large flight vehicles (rockets) is not only complex, but also can be very costly. These flights are infrequent and can lead to disastrous effects if something were to fail during the flight. The development of sensors coupled to internal components offers a great challenge in reducing their size, yet still maintaining their precision. Sounding rockets provide both a viable and convenient alternative to the more costly vehicular flights. Some of the major objectives are to test various types of sensors for monitoring components of high interest as well as investigating real-time processing techniques. Signal processing presents an extreme challenge in this noisy multichannel environment. The estimation and tracking of modal frequencies from vibrating structures is an important set of features that can provide information about the components under test; therefore, high resolution multichannel spectral processing is required. The application of both single channel and multichannel techniques capable of producing reliable modal frequency estimates of a vibrating structure from uncertain accelerometer measurements is discussed.

No abstract

No abstract