Publications

The flexibility offered by the Open Charge Point Protocol (OCPP) through the introduction of custom error codes also creates its own set of challenges. While the integration of custom error codes allows for enhanced granularity, it also introduces inconsistencies and fragmentation within the overarching diagnostic reporting system. Consequently, these custom error codes add an additional layer of complexity to the already intricate tasks of error reporting, diagnostics, and resolution. The variation in the definition of custom error codes makes it difficult to assess which entity in the charging ecosystem is responsible to correct errors and hinders the implementation of uniform error handling procedures across diverse charging stations and management systems. This results in prolonged resolution times and increased maintenance costs, resulting in decreased charging reliability. The challenge of charging reliability can be a significant obstacle to the widespread adoption of EVs, emphasizing the urgent need for a more robust approach to error handling across the EV charging ecosystem.

This paper proposes an Orthogonal Autoencoded Long-Short- Term Memory (OALSTM) network for long-term the State-Of-Charge (SOC) forecasting in Lithium-ion (Li-ion) battery cells. By leveraging the use of LSTMs in capturing temporal trends and orthogonal Autoencoder for extracting non- trivial robust latent features, OALSTM can achieve precise and accurate long-term SOC estimations near the end-of-life. One key contribution is learning orthogonal temporal encodings that generalize for long-term forecasting because it reduces the likelihood of false multicollinearity. Our results show that OALSTM outperforms other benchmark models for long-term SOC estimation of Li-ion battery cells under varying charging and discharging conditions.

Predictive analytics has emerged as a vital field with significant potential in industries ranging from energy to mobility. As such, it has become a topic of considerable interest for research. Through the use of statistical models, predictive analytics helps reveal patterns and relationships in complex datasets, generating accurate predictions about future events or outcomes. The development of Artificial Intelligence (AI) architectures and data-driven frameworks has further revolutionized the way we perform predictive analytics. However, the broad adoption of AI for predictive analytics is limited due to the lack of custom architectures that can effectively handle the unique complexities of modern datasets and perform robust and accurate predictions. As datasets grow increasingly complex, the need for Bayesian statistics and Deep Learning (DL) in predictive analytics has become increasingly evident. Bayesian statistics offers a versatile framework for incorporating prior knowledge and external knowledge into AI models. This can help mitigate problems such as data sparsity and improve long-term forecasts. Similarly, DL architectures, with their ability to identify and learn complex patterns within datasets, have the potential to unlock new insights and drive innovation in predictive analytics. However, the development of custom AI architectures that leverage such techniques for predictive analytics remains challenging due to their several inherent limitations. This work aims to bridge this research gap by harnessing the power of Bayesian statistics and DL to advance the state-of-the-art in predictive analytics. Specifically, this work proposes custom AI architectures and data-driven frameworks that can (i) perform accurate long-term estimations, (ii) overcome data drift, (iii) provide uncertainty quantifications, (iv) model and predict anomalous behavior, (v) leverage concepts of Design of Experiments, and (vi) perform collaborative modeling. The proposed models and frameworks are evaluated using compelling case studies that demonstrate their effectiveness in improving the accuracy, reliability, and robustness of AI architectures for broader use in predictive analytics.

For the safe and reliable operation of battery-driven machines, accurate state-of-charge (SOC) estimations are necessary. Unfortunately, existing methods often fail to identify patterns relevant to long-term SOC estimation due to complex battery cell characteristics such as aging. In this paper, we propose the Uncorrelated Sparse Autoencoder with Long Short-Term Memory (USAL). USAL is a novel neural network that addresses the challenging task of long-term SOC estimation given a limited initial history of a cell’s charge-discharge behavior. USAL uses a multi-task learning strategy to harness the advantages of sparse autoencoders and Long Short-term Memory (LSTM) networks by enforcing correlation penalties. The USAL simultaneously learns to (i) generate a latent space of informative SOC encodings from commonly measured cell characteristics, (ii) penalize for high multicollinearity between encodings, and (iii) identify non-trivial long and short temporal correlations between the encodings using LSTM cells. USAL outperforms benchmarked models in our experiments when trained on five initial charge-discharge cycles across multiple battery cells using three publicly available accelerated aging datasets. Note to Practitioners —This paper proposes USAL, a custom-built deep neural network to address the challenging task of long-term SOC estimations in battery cells. Long-term SOC estimation involves estimating SOC for cycles near End-Of-Life (EOL) given some initial charge-discharge cycles. Three fundamental steps involved in long-term SOC estimations using USAL are (i) exploiting a multi-task learning strategy to learn efficient encodings given limited training data, (ii) penalizing these encodings for high correlations to efficiently transform measured inputs into a space of informative features, and (iii) mapping of aging-related trends to support long-term SOC estimations. USAL is designed to be a data-driven SOC estimation method that is (i) capable of alerting the user to a faulty cell when integrated into a real-life Battery Management System (BMS) and (ii) identifying the relative quality of a battery cell from only a few initial charge-discharge cycles.

Although several state-of-charge (SOC) estimation methods have been proposed at the battery cell level, limited work has been done to identify the effect of cell aging on SOC estimations. To address this challenge, this paper proposes a novel method for estimating the SOC of Lithium-ion (Li-ion) battery cells by accurately modeling the cell aging and degradation information. The proposed method, termed as NNGP, is a deep neural network with Gaussian process feedback. The novel Gaussian process feedback helps the NNGP correlate SOC trends over consecutive charge-discharge cycles. Instead of time, the NNGP leverages available energy to correlate these SOC trends. The deep neural network within the NNGP then utilizes this information and other measured inputs to capture long-term cell aging and degradation trends. The NNGP leverages the most recent cell information to adapt its weights and estimate the SOC by employing an adaptive weighted training strategy. In our experiments on four Li-ion battery cells from three publicly available accelerated aging datasets, the NNGP clearly outperforms other benchmarked methods. The NNGP is also shown to be a useful prognostic tool capable of accurately estimating the SOC for up to 25 cycles in the future with an MAE below 3.5%. When tested under dynamic driving conditions and unknown initial SOC, the NNGP is shown to offer considerable improvements over other benchmarked state-of-art methods. The derivation of the model followed by experimental evaluation is presented.

Forecasting warranty claims for complex products is a reliability challenge for most manufacturers. Several factors increase the complexity of warranty claims forecasting, including, the limited number of claims reported at the early stage of launch, reporting delays, dynamic change in the fleet size, and design/manufacturing adjustments for the production line. The aggregated effect of those complexities is often referred to as the “warranty data maturation” effect. Unfortunately, most of the existing models for warranty claims forecasting fail to explicitly consider warranty data maturation. This work address warranty data maturation by proposing the Conditional Gaussian Mixture Model (CGMM). CGMM uses historical warranty data from similar products to develop a robust prior joint Gaussian mixture distribution of warranty trends at both, the current and future maturation levels. CGMM then utilizes Bayesian theories to estimate the conditional posterior distribution of the warranty claims at the future maturation level conditional on the warranty data available at the current maturation level. The CGMM identifies non-parametric temporal warranty trends and automatically clusters products into latent groups to establish (learn) an effective prior joint distribution. The CGMM is validated on an extensive automotive warranty claims dataset comprising of four model years and >15,000 different components from >10 million vehicles.

This paper proposes the Sparse Autoencoded Long Short-Term Memory network (SAEL) for long-term State-of-Charge (SOC) estimations. SAEL addresses the challenge of estimating the SOC near the end-of-life after only running a few charge-discharge cycles. SAEL transforms the inputs (e.g., voltage) into a space of informative features for SOC estimations. SAEL then feeds the transformed features into an LSTM network to identify temporal trends that support long-term SOC estimation. In our experiments, SAEL outperformed benchmark models by over 63% when evaluated on three battery cells. SAEL showed an MAE of 2.6% for the last twenty cycles when trained only on the initial five charge-discharge cycles.

This paper proposes a hybrid LSTM network for robust state-of-charge estimation of Li-ion batteries. The proposed model improves the estimation accuracy of a typical LSTM by using the SOC estimations of other trained machine learning (ML) models in addition to the original measurable battery cell parameters to train the LSTM. The hybrid LSTM intrinsically learns to timely activate the proper ML model by learning the complex dependencies between the accuracy of ML models and cell parameters. The proposed model is shown to achieve around 25% improvement in MAE for the last twenty cycles (near end-of-life) SOC estimation.

Over the last few decades, many architectures have been developed that harness the power of neural networks to detect objects in near real-time. Training such systems requires substantial time across multiple GPUs and massive labeled training datasets. Although the goal of these systems is generalizability, they are often impractical in real-life applications due to flexibility, robustness, or speed issues. This paper proposes RMOPP- A robust multi-objective post-processing algorithm to boost the performance of fast pre-trained object detectors with a negligible impact on their speed. Specifically, RMOPP is a statistically driven, post-processing algorithm that allows for simultaneous optimization of precision and recall. A unique feature of RMOPP is the Pareto frontier that identifies dominant possible post-processed detectors to optimize for both precision and recall. RMOPP explores the full potential of a pre-trained object detector and is deployable for near real-time predictions. We also provide a compelling test case on YOLOv2 using the MS-COCO dataset.

This paper proposes a new adaptive learning model for capacity estimation of lithium-ion battery cells. The proposed deep neural network transfers knowledge from other cells and adapts its behavior by exponentially weighting the data from the historical cells using a custom weighting function. The proposed model is shown to achieve state-of-art with an MAE of 0.56% when compared with three other traditional transfer learning and adaptive learning models for Li-ion battery cells. Details of the model followed by derivations and experimental results are provided.

This paper proposes a new long short-term memory neural network model to estimate the state-of-charge (SOC) of lithium-ion (Li-ion) battery cells. The proposed model improves the estimation accuracy by accounting for the changes in the battery parameters due to ageing by utilizing relevant knowledge from previous cycles when estimating the current state-of-charge. Derivation and details of the proposed model followed by experimental verification using commercial Li-ion battery cells are provided.

This paper proposes a neural network model for state-of-charge (SOC) estimation in lithium-ion battery cells. The proposed deep neural network model is a cycle-based recurrent model that leverages relevant information from historical cycles to provide reliable estimates of the state-of-charge of on-going cycles within a mean-absolute error (MAE) of 1%. In addition, the proposed model can be trained in a relatively short time. Details on the model followed by experimental verification are provided.

This paper presents a survey-cum-evaluation of methods for the comprehensive comparison of the task of keyword extraction using datasets of various sizes, forms, and genre. We use four different datasets which includes Amazon product data - Automotive, SemEval 2010, TMDB and Stack Exchange. Moreover, a subset of 100 Amazon product reviews is annotated and utilized for evaluation in this paper, to our knowledge, for the first time. Datasets are evaluated by five Natural Language Processing approaches (3 unsupervised and 2 supervised), which include TF-IDF, RAKE, TextRank, LDA and Shallow Neural Network. We use a ten-fold cross-validation scheme and evaluate the performance of the aforementioned approaches using recall, precision and F-score. Our analysis and results provide guidelines on the proper approaches to use for different types of datasets. Furthermore, our results indicate that certain approaches achieve improved performance with certain datasets due to inherent characteristics of the data.

The increase in the demand for electricity in commercial areas is going up by the day. The recent few years have shown a lot of growth in the commercial–express feeder consumers. Consequently, the need for energy is going up. Thus to satiate the demand, the utility company needs to produce more electricity, hence increasing the economics. To cope up with the rising cost and escalating demand of electricity we have come up with an indigenous solution. This solution aims at fulfilling electricity demand of the stated consumers by domestically generating electricity. Along with satisfying the demand for electricity of the consumer, a solution for exporting excessively generated electricity to the power grid has also been developed. The power thus generated is at a lower cost as compared to the commercial utility company tariff. This is accomplished with the use of micro steam turbine, its accessories and boiler. Conclusively providing a self-sufficient and profit path to the stated consumer.