Search for: All records

Abstract Electronic components that undergo shock and vibration are susceptible to failure caused by damage in the base printed circuit board that makes up the substrate of these systems. In certain applications, it may become paramount to know in real-time if the electronic components are damaged to enable a next-generation active system to take immediate responses. Broad examples of such systems include blast mitigation systems or safety systems in car accidents. These systems on classified under the term “high-rate” as they experience high shock levels on short time scales. This work proposes a long short-term memory neural network to enable real-time damage detection and assessment of electronic assemblies subjected to shock. The long short-term memory neural network is able to infer the state of the structure in approximately 4 milliseconds following the impact. The model obtains perfect classification results at 4 milliseconds for the data used in this work. This work is supported by experimentation that indicates damage to electronic packages can be quantified through the in situ monitoring of the impedance of electrical connections. Changes in impedance correlate to alterations in the physical properties of electronic components which indicate the occurrence of damage. On this basis, a comprehensive dataset is created to monitor the impedance changes of a daisy-chained connection through repeated high-energy shocks. Meanwhile, the shock response of the electronic components is captured using an accelerometer, enabling a detailed analysis of the effects of high-rate shock on the components’ performance. A dataset is developed to encompass 30 repeated impacts experiencing 10,000 gn during impact with an average half-sine time of 322 microseconds. The paper outlines the proposed real-time machine learning framework while performance metrics are presented and discussed in detail. 

Abstract To enable real-time control of next-generation active structures during shock events, there is a need to identify the start of a shock event within microseconds of its initiation. The delayed classification of a shock event may cause damage to the system that could have been prevented with assumed next-generation active control mechanisms. Addressing the challenge of ultra-low latency shock event classification requires utilizing prior information on normal behaviors (i.e., the system under vibrational loading) to identify abnormalities that can be classified as features of a shock event. The purpose of changepoint shock classification is to automatically recognize when a structure of interest behaves differently than expected in some measurable way. In this work, we analyze two different methods for shock classification using changepoint methodologies. We study the use of adaptive cumulative summation and expectation maximization algorithms in this work. Each method presents advantages and disadvantages for different scenarios. This study aims to derive features (streams of time series data) for the changepoint algorithms and revise the changepoint models to be used in real-time robust shock event detection. In this work, a printed circuit board under continuous vibrations before, during, and after a shock event is used to investigate the proposed methodologies. The printed circuit board is monitored with an accelerometer that is used to monitor both the vibrational and shock state of the system. The vibrational response of the system consists of accelerations up to 20 m/s2, while the shock event consists of loadings up to 2,000 m/s2. This work showed that the CUSUM algorithm is fairly effective at identifying the shock state in data but generates many false positives during normal behavior times, with no false positives post-shock, indicating accurate shock state detection despite early errors. In contrast, the Expectation Maximization (EM) algorithm shows improved performance by correctly predicting no shock in the initial phase and accurately identifying the onset of the shock state. It occasionally misclassifies shocked points as normal due to its change point identification process. Compared to CUSUM, EM has fewer false positives before the shock and similar performance during and after the shock event. Future research efforts will focus on developing online versions of these algorithms, which can identify system states with a minimum number of errors. The limitations of the system and its robustness to noise are discussed. 

Abstract Structures operating in high-rate dynamic environments, such as hypersonic vehicles, orbital space infrastructure, and blast mitigation systems, require microsecond (μs) decision-making. Advances in real-time sensing, edge-computing, and high-bandwidth computer memory are enabling emerging technologies such as High-rate structural health monitoring (HR-SHM) to become more feasible. Due to the time restrictions such systems operate under, a target of 1 millisecond (ms) from event detection to decision-making is set at the goal to enable HR-SHM. With minimizing latency in mind, a data-driven method that relies on time-series measurements processed in real-time to infer the state of the structure is investigated in this preliminary work. A methodology for deploying LSTM-based state estimators for structures using subsampled time-series vibration data is presented. The proposed estimator is deployed to an embedded real-time device and the achieved accuracy along with system timing are discussed. The proposed approach has shown potential for high-rate state estimation as it provides sufficient accuracy for the considered structure while a time-step of 2.5 ms is achieved. The Contributions of this work are twofold: 1) a framework for deploying LSTM models in real-time for high-rate state estimation, 2) an experimental validation of LSTMs running on a real-time computing system. 

Abstract In this paper, a method for real-time forecasting of the dynamics of structures experiencing nonstationary inputs is described. This is presented as time series predictions across different timescales. The target applications include hypersonic vehicles, space launch systems, real-time prognostics, and monitoring of high-rate and energetic systems. This work presents numerical analysis and experimental results for the real-time implementation of a Fast Fourier Transform (FFT)-based approach for time series forecasting. For this preliminary study, a testbench structure that consists of a cantilever beam subjected to nonstationary inputs is used to generate experimental data. First, the data is de-trended, then the time series data is transferred into the frequency domain, and measures for frequency, amplitude, and phase are obtained. Thereafter, select frequency components are collected, transformed back to the time domain, recombined, and then the trend in the data is restored. Finally, the recombined signals are propagated into the future to the selected prediction horizon. This preliminary time series forecasting work is done offline using pre-recorded experimental data, and the FFT-based approach is implemented in a rolling window configuration. Here learning windows of 0.1, 0.5, and 1 s are considered with different computation times simulated. Results demonstrate that the proposed FFT-based approach can maintain a constant prediction horizon at 1 s with sufficient accuracy for the considered system. The performance of the system is quantified using a variety of metrics. Computational speed and prediction accuracy as a function of training time and learning window lengths are examined in this work. The algorithm configuration with the shortest learning window (0.1 s) is shown to converge faster following the nonstationary when compared to algorithm configuration with longer learning windows. 

Deep-Learning has become a dominant computing paradigm across a broad range of application domains. Different architectures of Deep-Networks like CNN, MLP, and RNN have emerged as the prominent machine-learning approaches for today’s application domains. These architectures are heavily data-dependent, requiring frequent access to memory. As a result, these applications suffer the most from the memory bottleneck of the von Neumann architectures. There is an imminent need for memory-centric architectures for deep-learning and big-data analytic applications that are memory intensive. Modern Field Programmable Gate Arrays (FPGAs) are ideal programmable substrates for creating customized Processor in/near Memory (PIM) accelerators. Modern FPGAs contain 100s of Mbits of dual-ported SRAM in the form of disaggregated, configurable Block RAMs (BRAMs). These BRAMs contain TB/s of available internal bandwidth. Unfortunately, developing FPGA-based accelerators for deep learning is not a simple task and demands the utilization of specialized tools provided by the FPGA vendors. It requires expertise in low-level hardware microarchitecture design. These are often not available to most researchers in the field of deep learning. Even with the ongoing improvements in High-Level Synthesis (HLS) tools, the requirement for hardware-specific design knowledge cannot be completely eliminated. This research developed a new reconfigurable memory-centric architecture and design approach that opens the advantages of FPGAs and Processor-in-Memory architecture to memory-intensive applications. Due to its high-performance and scalable memory-centric design, this architecture can deliver the highest speed and the lowest latency achievable from an FPGA overcoming the memory bottleneck.