An official website of the United States government
100% inverter-based renewable units are becoming more prevalent, introducing new challenges in the protection of microgrids that incorporate these resources. This is particularly due to low fault currents and bidirectional flows. Previous work has studied the protection of microgrids with high penetration of inverter-interfaced distributed generators; however, very few have studied the protection of a 100% inverter-based microgrid. This work proposes machine learning (ML)–based protection solutions using local electrical measurements that consider implementation challenges and effectively combine short-circuit fault detection and type identification. A decision tree method is used to analyze a wide range of fault scenarios. PSCAD/EMTDC simulation environment is used to create a dataset for training and testing the proposed method. The effectiveness of the proposed methods is examined under seven distinct fault types, each featuring varying fault resistance, in a 100% inverter-based microgrid consisting of four inverters.
more »
« less
Comparison of Manifold Learning Algorithms for Rapid Circuit Defect Extraction in SPICE-Augmented Machine Learning
Identifying the source of integrated circuit (IC) degradation and being able to track its degradation via its electrical characteristics (e.g. the Voltage Transfer Characteristics, VTC, of an inverter) is very useful in failure analysis. This is because the electrical measurement is non-destructive, low-cost, and rapid. However, the extraction of defects from electrical characteristics requires significant domain expertise. To reduce or even obviate the need for domain expertise so that the process can be automatic for various circuits, one may use manifold learning. As a type of machine learning (ML), manifold learning also requires a large amount of accurate training data. To obtain enough defect training data, which is almost impossible from experiments, one may use SPICE simulation. Based on our previous work of using AutoEncoder (AE) to perform SPICE-augmented ML to extract the pMOS and nMOS source contact resistances from the inverter VTC, in this paper, we compare the efficacy of using another 6 types of manifold learning. They are used to predict the experimental result and it is found that most of them have reasonable performance although the AE is still the best (R2=0.9). However, when including also the variation of PMOS width (as a weak perturbation to the data), algorithms such as Locally Linear Embedding (LLE) are found to perform better than AE (R2=0.72) with LLE (R2=0.83) being the best. Therefore, multiple manifold learnings are suggested to be used in parallel in real production to enhance accuracy.
more »
« less
- Award ID(s):
- 2046220
- PAR ID:
- 10324370
- Date Published:
- Journal Name:
- 2022 IEEE 19th Annual Workshop on Microelectronics and Electron Devices (WMED), 2022
- Page Range / eLocation ID:
- 1 to 4
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Optical network failure management (ONFM) is a promising application of machine learning (ML) to optical networking. Typical ML-based ONFM approaches exploit historical monitored data, retrieved in a specific domain (e.g., a link or a network), to train supervised ML models and learn failure characteristics (a signature) that will be helpful upon future failure occurrence in that domain. Unfortunately, in operational networks, data availability often constitutes a practical limitation to the deployment of ML-based ONFM solutions, due to scarce availability of labeled data comprehensively modeling all possible failure types. One could purposely inject failures to collect training data, but this is time consuming and not desirable by operators. A possible solution is transfer learning (TL), i.e., training ML models on a source domain (SD), e.g., a laboratory testbed, and then deploying trained models on a target domain (TD), e.g., an operator network, possibly fine-tuning the learned models by re-training with few TD data. Moreover, in those cases when TL re-training is not successful (e.g., due to the intrinsic difference in SD and TD), another solution is domain adaptation, which consists of combining unlabeled SD and TD data before model training. We investigate domain adaptation and TL for failure detection and failure-cause identification across different lightpaths leveraging real optical SNR data. We find that for the considered scenarios, up to 20% points of accuracy increase can be obtained with domain adaptation for failure detection, while for failure-cause identification, only combining domain adaptation with model re-training provides significant benefit, reaching 4%–5% points of accuracy increase in the considered cases.more » « less
-
One of the goals of undergraduate education is to prepare students to adapt to a challenging career that requires continual learning and application of knowledge. Working professionals should have deep conceptual knowledge that they can apply in a range of contexts and possess the attitudes and skills of lifelong learners. The literature suggests the concept of Adaptive Expertise (AE), which can be defined as the ability to apply and extend knowledge and skills to new situations, describes some of these characteristics. Survey data concerning the level of AE displayed by various populations is extremely limited in most contexts, be it education or working professionals. As such, data concerning the level of adaptiveness displayed among various groups needs to be measured if activities designed to promote the development of AE are to be created and then tested in terms of their efficacy. This investigation provides this critical baseline data for future studies as we track the AE development of individual, first-year college students through their undergraduate program of study, with a focus on low-income students as a means to support retention. In this work, we assessed adaptive expertise among low-income STEM students using surveys and interviews. Low-income STEM students from various stages of their four year undergraduate program (n=208) completed an adaptive expertise survey in spring 2022. Following the survey, 24 of the low-income students (6 per year, 3 male, 3 female) were selected for targeted qualitative interviews to better understand the differences displayed by low and high AE students. Survey results from prior studies were used to draw comparisons between adaptiveness of low-income and non-low-income students. Results of the AE survey indicated no statistically-significant differences between low-income and non-low-income first-year students in terms of their level of adaptiveness. In addition, the level of AE displayed by low-income students increased through the program in a manner similar to that of non-low-income STEM students. Themes that emerged from the interviews included a general understanding of the importance and likelihood of learning new concepts continually while working in a professional role, and that students expressed growth in understanding the acceptance of reaching out for assistance from other students and faculty after exploring information on their own as they work through challenges in their academic assignments. Two dominant and divergent metacognitive processes were also observed: teaching/explaining concepts to others (highly adaptive) or primarily relying on exam/course grades for feedback on learning (low adaptiveness). Data gathered from interviews demonstrate the need for a greater emphasis on metacognitive practice to promote various aspects of AE.more » « less
-
Abstract Modern machine learning (ML) and deep learning (DL) techniques using high-dimensional data representations have helped accelerate the materials discovery process by efficiently detecting hidden patterns in existing datasets and linking input representations to output properties for a better understanding of the scientific phenomenon. While a deep neural network comprised of fully connected layers has been widely used for materials property prediction, simply creating a deeper model with a large number of layers often faces with vanishing gradient problem, causing a degradation in the performance, thereby limiting usage. In this paper, we study and propose architectural principles to address the question of improving the performance of model training and inference under fixed parametric constraints. Here, we present a general deep-learning framework based on branched residual learning (BRNet) with fully connected layers that can work with any numerical vector-based representation as input to build accurate models to predict materials properties. We perform model training for materials properties using numerical vectors representing different composition-based attributes of the respective materials and compare the performance of the proposed models against traditional ML and existing DL architectures. We find that the proposed models are significantly more accurate than the ML/DL models for all data sizes by using different composition-based attributes as input. Further, branched learning requires fewer parameters and results in faster model training due to better convergence during the training phase than existing neural networks, thereby efficiently building accurate models for predicting materials properties.more » « less
-
null (Ed.)Weakly labeled data are inevitable in various research areas in artificial intelligence (AI) where one has a modicum of knowledge about the complete dataset. One of the reasons for weakly labeled data in AI is insufficient accurately labeled data. Strict privacy control or accidental loss may also cause missing-data problems. However, supervised machine learning (ML) requires accurately labeled data in order to successfully solve a problem. Data labeling is difficult and time-consuming as it requires manual work, perfect results, and sometimes human experts to be involved (e.g., medical labeled data). In contrast, unlabeled data are inexpensive and easily available. Due to there not being enough labeled training data, researchers sometimes only obtain one or few data points per category or label. Training a supervised ML model from the small set of labeled data is a challenging task. The objective of this research is to recover missing labels from the dataset using state-of-the-art ML techniques using a semisupervised ML approach. In this work, a novel convolutional neural network-based framework is trained with a few instances of a class to perform metric learning. The dataset is then converted into a graph signal, which is recovered using a recover algorithm (RA) in graph Fourier transform. The proposed approach was evaluated on a Fashion dataset for accuracy and precision and performed significantly better than graph neural networks and other state-of-the-art methodsmore » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/WMED55302.2022.9758032
