Dissertations / Theses: 'Machine learning not elsewhere classified'

1

Abo, Al Ahad George, and Abbas Salami. "Machine Learning for Market Prediction : Soft Margin Classifiers for Predicting the Sign of Return on Financial Assets." Thesis, Linköpings universitet, Produktionsekonomi, 2018. http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-151459.

Full text

Abstract:

Forecasting procedures have found applications in a wide variety of areas within finance and have further shown to be one of the most challenging areas of finance. Having an immense variety of economic data, stakeholders aim to understand the current and future state of the market. Since it is hard for a human to make sense out of large amounts of data, different modeling techniques have been applied to extract useful information from financial databases, where machine learning techniques are among the most recent modeling techniques. Binary classifiers such as Support Vector Machines (SVMs) have to some extent been used for this purpose where extensions of the algorithm have been developed with increased prediction performance as the main goal. The objective of this study has been to develop a process for improving the performance when predicting the sign of return of financial time series with soft margin classifiers. An analysis regarding the algorithms is presented in this study followed by a description of the methodology that has been utilized. The developed process containing some of the presented soft margin classifiers, and other aspects of kernel methods such as Multiple Kernel Learning have shown pleasant results over the long term, in which the capability of capturing different market conditions have been shown to improve with the incorporation of different models and kernels, instead of only a single one. However, the results are mostly congruent with earlier studies in this field. Furthermore, two research questions have been answered where the complexity regarding the kernel functions that are used by the SVM have been studied and the robustness of the process as a whole. Complexity refers to achieving more complex feature maps through combining kernels by either adding, multiplying or functionally transforming them. It is not concluded that an increased complexity leads to a consistent improvement, however, the combined kernel function is superior during some of the periods of the time series used in this thesis for the individual models. The robustness has been investigated for different signal-to-noise ratio where it has been observed that windows with previously poor performance are more exposed to noise impact.

APA, Harvard, Vancouver, ISO, and other styles

2

(10506350), Amogh Agrawal. "Compute-in-Memory Primitives for Energy-Efficient Machine Learning." Thesis, 2021.

Find full text

Abstract:

Machine Learning (ML) workloads, being memory and compute-intensive, consume large amounts of power running on conventional computing systems, restricting their implementations to large-scale data centers. Thus, there is a need for building domain-specific hardware primitives for energy-efficient ML processing at the edge. One such approach is in-memory computing, which eliminates frequent and unnecessary data-transfers between the memory and the compute units, by directly computing the data where it is stored. Most of the chip area is consumed by on-chip SRAMs in both conventional von-Neumann systems (e.g. CPU/GPU) as well as application-specific ICs (e.g. TPU). Thus, we propose various circuit techniques to enable a range of computations such as bitwise Boolean and arithmetic computations, binary convolution operations, non-Boolean dot-product operations, lookup-table based computations, and spiking neural network implementation - all within standard SRAM memory arrays.

First, we propose X-SRAM, where, by using skewed sense amplifiers, bitwise Boolean operations such as NAND/NOR/XOR/IMP etc. can be enabled within 6T and 8T SRAM arrays. Moreover, exploiting the decoupled read/write ports in 8T SRAMs, we propose read-compute-store scheme where the computed data can directly be written back in the array simultaneously.

Second, we propose Xcel-RAM, where we show how binary convolutions can be enabled in 10T SRAM arrays for accelerating binary neural networks. We present charge sharing approach for performing XNOR operations followed by a population count (popcount) using both analog and digital techniques, highlighting the accuracy-energy tradeoff.

Third, we take this concept further and propose CASH-RAM, to accelerate non-Boolean operations, such as dot-products within standard 8T-SRAM arrays by utilizing the parasitic capacitances of bitlines and sourcelines. We analyze the non-idealities that arise due to analog computations and propose a self-compensation technique which reduces the effects of non-idealities, thereby reducing the errors.

Fourth, we propose ROM-embedded caches, RECache, using standard 8T SRAMs, useful for lookup-table (LUT) based computations. We show that just by adding an extra word-line (WL) or a source-line (SL), the same bit-cell can store a ROM bit, as well as the usual RAM bit, while maintaining the performance and area-efficiency, thereby doubling the memory density. Further we propose SPARE, an in-memory, distributed processing architecture built on RECache, for accelerating spiking neural networks (SNNs), which often require high-order polynomials and transcendental functions for solving complex neuro-synaptic models.

Finally, we propose IMPULSE, a 10T-SRAM compute-in-memory (CIM) macro, specifically designed for state-of-the-art SNN inference. The inherent dynamics of the neuron membrane potential in SNNs allows processing of sequential learning tasks, avoiding the complexity of recurrent neural networks. The highly-sparse spike-based computations in such spatio-temporal data can be leveraged for energy-efficiency. However, the membrane potential incurs additional memory access bottlenecks in current SNN hardware. IMPULSE triew to tackle the above challenges. It consists of a fused weight (WMEM) and membrane potential (VMEM) memory and inherently exploits sparsity in input spikes. We propose staggered data mapping and re-configurable peripherals for handling different bit-precision requirements of WMEM and VMEM, while supporting multiple neuron functionalities. The proposed macro was fabricated in 65nm CMOS technology. We demonstrate a sentiment classification task from the IMDB dataset of movie reviews and show that the SNN achieves competitive accuracy with only a fraction of trainable parameters and effective operations compared to an LSTM network.

These circuit explorations to embed computations in standard memory structures shows that on-chip SRAMs can do much more than just store data and can be re-purposed as on-demand accelerators for a variety of applications.

APA, Harvard, Vancouver, ISO, and other styles

3

(9189263), Shuo Han. "FLUORESCENCE MICROSCOPY IMAGES SEGMENTATION AND ANALYSIS USING MACHINE LEARNING." Thesis, 2020.

Find full text

Abstract:

Microscopy image analysis can provide substantial information for clinical study and understanding of the biological structure. Two-photon microscopy is a type of fluorescence microscopy that can visualize deep into tissue with near-infrared excitation light. Large 3D image volumes of complex subcellular are often produced, which calls for automatic image analysis techniques. Automatic methods that can obtain nuclei quantity in microscopy image volumes are needed for biomedical research and clinical diagnosis. In general, several challenges exist for counting nuclei in 3D image volumes. These include “crowding” and touching of nuclei, overlapping of two or morenuclei, and shape and size variances of the nuclei. In this thesis, a 3D nuclei counterusing two different generative adversarial networks (GAN) is proposed and evaluated.Synthetic data that resembles real microscopy image is generated with a GAN. The synthetic data is used to train another 3D GAN network that counts the number o fnuclei. Our approach is evaluated with respect to the number of ground truth nuclei and compared with common ways of counting used in the biological research.Fluorescence microscopy 3D image volumes of rat kidneys are used to test our 3D nuclei counter. The evaluation of both networks shows that the proposed technique is successful for counting nuclei in 3D. Then, a 3D segmentation and classification method to segment and identify individual nuclei in fluorescence microscopy volumes without having ground truth volumes is introduced. Three dimensional synthetic data is generated using the Recycle-GAN with the Hausdorff distance loss introduced into preserve the shape of individual nuclei. Realistic microscopy image volumes with nuclei segmentation mask and nucleus boundary ground truth volumes are generated.A subsequent 3D CNN with a regularization term that discourage detection out of nucleus boundary is used to detect and segment nuclei. Nuclei boundary refinement is then performed to enhance nuclei segmentation. Experimental results on our rat kidney dataset show the proposed method is competitive with respect to several state-of-the-art methods. A Distributed and Networked Analysis of Volumetric Image Data(DINAVID) system is developed to enable remote analysis of microscopy images for biologists. There are two main functions integrated in the system, a 3D visualization tool and a remote computing tool for nuclei segmentation. The 3D visualization enables real-time rendering of large volumes of microscopy data. The segmentation tool provides fast inferencing of pre-trained deep learning models trained with 5 different types of microscopy data.

APA, Harvard, Vancouver, ISO, and other styles

4

(5930189), Javier Ribera Prat. "Image-based Plant Phenotyping Using Machine Learning." Thesis, 2019.

Find full text

Abstract:

Phenotypic data is of crucial importance for plant breeding in estimating a plant's biomass. Traits such as leaf area and plant height are known to be correlated with biomass. Image analysis and computer vision methods can automate data analysis for high-throughput phenotyping. Many methods have been proposed for plant phenotyping in controlled environments such as greenhouses. In this thesis, we present multiple methods to estimate traits of the plant crop sorghum from images acquired from UAV and field-based sensors. We describe machine learning techniques to extract the plots of a crop field, a method for leaf counting from low-resolution images, and a statistical model that uses prior information about the field structure to estimate the center of each plant. We also develop a new loss function to train Convolutional Neural Networks (CNNs) to count and locate objects of any type and use it to estimate plant centers. Our methods are evaluated with ground truth of sorghum fields and publicly available datasets and are shown to outperform the state of the art in generic object detection and domain-specific tasks.

This thesis also examines the use of crowdsourcing information in video analytics. The large number of cameras deployed for public safety surveillance systems requires intelligent processing capable of automatically analyzing video in real time. We incorporate crowdsourcing in an online basis to improve a crowdflow estimation method. We present various approaches to characterize this uncertainty and to aggregate crowdsourcing results. Our techniques are evaluated using publicly available datasets.

APA, Harvard, Vancouver, ISO, and other styles

5

(7534550), David Güera. "Media Forensics Using Machine Learning Approaches." Thesis, 2019.

Find full text

Abstract:

Consumer-grade imaging sensors have become ubiquitous in the past decade. Images and videos, collected from such sensors are used by many entities for public and private communications, including publicity, advocacy, disinformation, and deception.

In this thesis, we present tools to be able to extract knowledge from and understand this imagery and its provenance. Many images and videos are modified and/or manipulated prior to their public release. We also propose a set of forensics and counter-forensic techniques to determine the integrity of this multimedia content and modify it in specific ways to deceive adversaries. The presented tools are evaluated using publicly available datasets and independently organized challenges.

APA, Harvard, Vancouver, ISO, and other styles

6

(9136835), Sungbum Jun. "SCHEDULING AND CONTROL WITH MACHINE LEARNING IN MANUFACTURING SYSTEMS." Thesis, 2020.

Find full text

Abstract:

Numerous optimization problems in production systems can be considered as decision-making processes that determine the best allocation of resources to tasks over time to optimize one or more objectives in concert with big data. Among the optimization problems, production scheduling and routing of robots for material handling are becoming more important due to their impacts on system performance. However, the development of efficient algorithms for scheduling or routing faces several challenges. While the scheduling and vehicle routing problems can be solved by mathematical models such as mixed-integer linear programming to find optimal solutions to smallsized problems, they are not applicable to larger problems due to the nature of NP-hard problems. Thus, further research on machine learning applications to those problems is a significant step towards increasing the possibilities and potentialities of field application. In order to create truly intelligent systems, new frameworks for scheduling and routing are proposed to utilize machine learning (ML) techniques. First, the dynamic single-machine scheduling problem for minimization of total weighted tardiness is addressed. In order to solve the problem more efficiently, a decisiontree-based approach called Generation of Rules Automatically with Feature construction and Treebased learning (GRAFT) is designed to extract dispatching rules from existing or good schedules. In addition to the single-machine scheduling problem, the flexible job-shop scheduling problem with release times for minimizing the total weighted tardiness is analyzed. As a ML-based solution approach, a random-forest-based approach called Random Forest for Obtaining Rules for Scheduling (RANFORS) is developed to solve the problem by generating dispatching rules automatically. Finally, an optimization problem for routing of autonomous robots for minimizing total tardiness of transportation requests is analyzed by decomposing it into three sub-problems. In order to solve the sub-problems, a comprehensive framework with consideration of conflicts between routes is proposed. Especially to the sub-problem for vehicle routing, a new local search algorithm called COntextual-Bandit-based Adaptive Local search with Tree-based regression (COBALT) that incorporates the contextual bandit into operator selection is developed. The findings from my research contribute to suggesting a guidance to practitioners for the applications of ML to scheduling and control problems, and ultimately to lead the implementation of smart factories.

APA, Harvard, Vancouver, ISO, and other styles

7

(8812160), Alex Joseph Raynor. "DEVELOPMENT OF MACHINE LEARNING TECHNIQUES FOR APPLICATIONS IN THE STEEL INDUSTRY." Thesis, 2020.

Find full text

Abstract:

For a long time, the collection of data through sensors and other means was seen as inconsequential. However, with the somewhat recent developments in the areas of machine learning, data science, and statistical analysis, as well as in the rapid growth of computational power being allotted by the ever-expanding computer industry, data is not just being seen as secondhand information anymore. Data collection is showing that it currently is and will continue to be a major driving force in many applications, as the predictive power it can provide is invaluable. One such area that could benefit dramatically from the use of predictive techniques is the steel industry. This thesis applied several machine learning techniques to predict steel deformation issues collectively known as the hook index problem [1].

The first machine learning technique utilized in this endeavor was neural networking. The neural networks built and tested in this research saw the use of classification and regression prediction models. They also implemented the algorithms of gradient descent and adaptive moment estimation. Through the employment of these networks and learning strategies, as well as through the line process data, regression-based networks made predictions with average percent error ranging from 106-114%. In similar performance to the regression-based networks, classification-based networks made predictions with average accuracy percentage ranges of 38-40%.

To remedy the problems relating to neural networks, Bayesian networking techniques were implemented. The main method that was used as a model for these networks was the Naïve Bayesian framework. Also, variable optimization techniques were utilized to create well-performing network structures. In the same vein as the neural networks, Bayesian networks used line process data to make predictions. The classification-based networks made predictions with average accuracy ranges of 64-65%. Because of the increased accuracy results and their ability to draw causal reasoning from data, Bayesian networking was the preferred machine learning technique for this research application.

APA, Harvard, Vancouver, ISO, and other styles

8

(9811085), Anand Koirala. "Precision agriculture: Exploration of machine learning approaches for assessing mango crop quantity." Thesis, 2020. https://figshare.com/articles/thesis/Precision_agriculture_Exploration_of_machine_learning_approaches_for_assessing_mango_crop_quantity/13411625.

Full text

Abstract:

A machine vision based system is proposed to replace the current in-orchard manual estimates of mango fruit yield, to inform harvest resourcing and marketing. The state-of-the-art in fruit detection was reviewed, highlighting the recent move from traditional image segmentation methods to convolution neural network (CNN) based deep learning methods. An experimental comparison of several deep learning based object detection frameworks (single shot detectors versus two-staged detectors) and several standard CNN architectures was undertaken for detection of mango panicles and fruit in tree images. The machine vision system used images of individual trees captured during night time from a moving platform mounted with a Global Navigation Satellite System (GNSS) receiver and a LED panel floodlight. YOLO, a single shot object detection framework, was re-designed and named as MangoYOLO. MangoYOLO outperformed existing state-of-the-art deep learning object detection frameworks in terms of fruit detection time and accuracy and was robust in use across different cultivars and cameras. MangoYOLO achieved F1 score of 0.968 and average precision of 0.983 and required just 70 ms per image (2048 × 2048 pixel) and 4417 MB memory. The annotated image dataset was made publicly available. Approaches were trialled to relate the fruit counts from tree images to the actual harvest count at an individual tree level. Machine vision based estimates of fruit load ranged between -11% to +14% of packhouse fruit counts. However, estimation of fruit yield (t/ha) requires estimation of fruit size as well as fruit number. A fruit sizing app for smart phones was developed as an affordable in-field solution. The solution was based on segmentation of the fruit in image using colour features and estimation of the camera to fruit perimeter distance based on use of fruit allometrics. For mango fruit, RMSEs of 5.3 and 3.7 mm were achieved on length and width measurements under controlled lighting, and RMSEs of 5.5 and 4.6 mm were obtained in-field under ambient lighting. Further, estimation of harvest timing can be informed by assessment of the spread of flowering. Deep learning object detection methods were deployed for assessment of the number and development stage of mango panicles, on tree. Methods to deal with different orientations of flower panicles in tree images were implemented. An R2 >0.8 was achieved between machine vision count of panicles on images and in-field human count per tree. Similarly, mean average precision of 69.1% was achieved for classification of panicle stages. These machine vision systems form a foundation for estimation of crop load and harvest timing, and for automated harvesting.

APA, Harvard, Vancouver, ISO, and other styles

9

(11173365), Youlin Liu. "MACHINE LEARNING METHODS FOR SPECTRAL ANALYSIS." Thesis, 2021.

Find full text

Abstract:

Measurement science has seen fast growth of data in both volume and complexity in recent years, new algorithms and methodologies have been developed to aid the decision
making in measurement sciences, and this process is automated for the liberation of labor. In light of the adversarial approaches shown in digital image processing, Chapter 2 demonstrate how the same attack is possible with spectroscopic data. Chapter 3 takes the question presented in Chapter 2 and optimized the classifier through an iterative approach. The optimized LDA was cross-validated and compared with other standard chemometrics methods, the application was extended to bi-distribution mineral Raman data. Chapter 4 focused on a novel Artificial Neural Network structure design with diffusion measurements; the architecture was tested both with simulated dataset and experimental dataset. Chapter 5 presents the construction of a novel infrared hyperspectral microscope for complex chemical compound classification, with detailed discussion in the segmentation of the images and choice of a classifier to choose.

APA, Harvard, Vancouver, ISO, and other styles

10

(10713342), Judy Yang. "Development of a warehouse model using machine learning technologies with application in receiving management." Thesis, 2021. https://figshare.com/articles/thesis/Development_of_a_warehouse_model_using_machine_learning_technologies_with_application_in_receiving_management/14499078.

Full text

Abstract:

Warehouse management is a significant element of a supply chain. The warehouse's data accuracy is the foundation of the whole supply chain material flow, information flow, and capital flow. It requires the warehouse receiving management to have an accurate system that operates in real-time to collect receiving data. However, conventional solutions for data management at the receiving stage are managed manually. Human involvement in the current receiving stage means that data accuracy and real-time entry cannot be assured. This research aims to improve data accuracy and reduce human error in the warehouse receiving management. The purpose of this research is to develop a suitable Artificial Neural Network (ANN) model by using machine learning technologies, which can be applied to warehouse receiving management to improve data accuracy and reduce human errors. Based on comprehensive literature review and analysis of current algorithms, a conceptual ANN model, identified as the Artificial Neural Network for Components Identification and Counting (ANN-CIC) model, is proposed to perform components classification and counting. This study has evaluated three classic image identification algorithms by using sixteen groups of industrial components to compare classification performance. A modified white histogram correlation coefficient approach is chosen as the design model's classification algorithm after experiments. Besides, the counting model is tested. The model is verified with an enlarged dataset obtained from a local Australian Company. The simulation results demonstrated that the proposed model achieved a 91.37% accuracy rate in object classification and a 94.29% in object counting, which has outperformed the existing classical model accuracy rate, such as for VGG-16. The main contributions of this research can be highlighted as below: Firstly, a conceptual ANN-CIC model is proposed to perform the identification and counting of industrial components. Four basic geometric shapes as the attributes of images for shape analysis and pre-defined features are introduced. These introduced shapes assisted in verifying the feasibility of the preliminary experiments. Secondly, the white histogram correlation coefficient algorithm is improved by adjusting the colour ratio to achieve outstanding performance in the classification of various industrial components. Lastly, the model is simulated with industrial data that demonstrates its applicability and stability. Moreover, higher classification and counting accuracy rate are achieved, and the design goal is also achieved.

APA, Harvard, Vancouver, ISO, and other styles

11

(12214559), Sonal Chawda. "Determination of distance relay characteristics using an inductive learning system." Thesis, 1993. https://figshare.com/articles/thesis/Determination_of_distance_relay_characteristics_using_an_inductive_learning_system/19326599.

Full text

Abstract:

In this research an attempt has been made to design distance relays as per protection system requirements. This is achieved by using an Inductive learning technique. The inductive learning algorithm which belongs to the family of machine learning by examples is used to convert a set of impedance values into a decision tree. The impedance values are obtained by conducting fault study on the system to be protected. A number of tests have been carried out on various transmission line configurations. The required software for generating the

decision tree has been developed.

APA, Harvard, Vancouver, ISO, and other styles

12

(11180610), Indranil Chakraborty. "Toward Energy-Efficient Machine Learning: Algorithms and Analog Compute-In-Memory Hardware." Thesis, 2021.

Find full text

Abstract:

The ‘Internet of Things’ has increased the demand for artificial intelligence (AI)-based edge computing in applications ranging from healthcare monitoring systems to autonomous vehicles. However, the growing complexity of machine learning workloads requires rethinking to make AI amenable to resource constrained environments such as edge devices. To that effect, the entire stack of machine learning, from algorithms to hardware primitives, have been explored to enable energy-efficient intelligence at the edge.

From the algorithmic aspect, model compression techniques such as quantization are powerful tools to address the growing computational cost of ML workloads. However, quantization, particularly, can result in substantial loss of performance for complex image classification tasks. To address this, a principal component analysis (PCA)-driven methodology to identify the important layers of a binary network, and design mixed-precision networks. The proposed Hybrid-Net achieves a significant improvement in classification accuracy over binary networks such as XNOR-Net for ResNet and VGG architectures on CIFAR-100 and ImageNet datasets, while still achieving up remarkable energy-efficiency.

Having explored compressed neural networks, there is a need to investigate suitable computing systems to further the energy efficiency. Memristive crossbars have been extensively explored as an alternative to traditional CMOS based systems for deep learning accelerators due to their high on-chip storage density and efficient Matrix Vector Multiplication (MVM) compared to digital CMOS. However, the analog nature of computing poses significant issues due to various non-idealities such as: parasitic resistances, non-linear I-V characteristics of the memristor device etc. To address this, a simplified equation-based modelling of the non-ideal behavior of crossbars is performed and correspondingly, a modified technology aware training algorithm is proposed. Building on the drawbacks of equation-based modeling, a Generalized Approach to Emulating Non-Ideality in Memristive Crossbars using Neural Networks (GENIEx) is proposed where a neural network is trained on HSPICE simulation data to learn the transfer characteristics of the non-ideal crossbar. Next, a functional simulator was developed which includes key architectural facets such as tiling, and bit-slicing to analyze the impact of non-idealities on the classification accuracy of large-scale neural networks.

To truly realize the benefits of hardware primitives and the algorithms on top of the stack, it is necessary to build efficient devices that mimic the behavior of the fundamental units of a neural network, namely, neurons and synapses. However, efforts have largely been invested in implementations in the electrical domain with potential limitations of switching speed, functional errors due to analog computing, etc. As an alternative, a purely photonic operation of an Integrate-and-Fire Spiking neuron is proposed, based on the phase change dynamics of Ge2Sb2Te5 (GST) embedded on top of a microring resonator, which alleviates the energy constraints of PCMs in electrical domain. Further, the inherent parallelism of wavelength-division multiplexing (WDM) was leveraged to propose a photonic dot-product engine. The proposed computing platform was used to emulate a SNN inferencing engine for image-classification tasks. These explorations at different levels of the stack can enable energy-efficient machine learning for edge intelligence.

Having explored various domains to design efficient DNN models and studying various hardware primitives based on emerging technologies, we focus on Silicon implementation of compute-in-memory (CIM) primitives for machine learning acceleration based on the more available CMOS technology. CIM primitives enable efficient matrix-vector multiplications (MVM) through parallelized multiply-and-accumulate operations inside the memory array itself. As CIM primitives deploy bit-serial computing, the computations are exposed bit-level sparsity of inputs and weights in a ML model. To that effect, we present an energy-efficient sparsity-aware reconfigurable-precision compute-in-memory (CIM) 8T-SRAM macro for machine learning (ML) applications. Standard 8T-SRAM arrays are re-purposed to enable MAC operations using selective current flow through the read-port transistors. The proposed macro dynamically leverages workload sparsity by reconfiguring the output precision in the peripheral circuitry without degrading application accuracy. Specifically, we propose a new energy-efficient reconfigurable-precision SAR ADC design with the ability to form (n+m)-bit precision using n-bit and m-bit ADCs. Additionally, the transimpedance amplifier (TIA) –required to convert the summed current into voltage before conversion—is reconfigured based on sparsity to improve sense margin at lower output precision. The proposed macro, fabricated in 65 nm technology, provides 35.5-127.2 TOPS/W as the ADC precision varies from 6-bit to 2-bit, respectively. Building on top of the fabricated macro, we next design a hierarchical CIM core micro-architecture that addresses the existing CIM scaling challenges. The proposed CIM core micro-architecture consists of 32 proposed sparsity-aware CIM macros. The 32 macros are divided into 4 matrix-vector multiplication units (MVMUs) consisting of 8 macros each. The core has three unique features: i) it can adaptively reconfigure ADC precision to achieve energy-efficiency and lower latency based on input and weight sparsity, determined by a sparsity controller, ii) it deploys row-gating feature to maintain SNR requirements for accurate DNN computations, and iii) hardware support for load balancing to balance latency mismatches occurring due to different ADC precisions in different compute units. Besides the CIM macros, the core micro-architecture consists of input, weight, and output memories, along with instruction memory and control circuits. The instruction set architecture allows for flexible dataflows and mapping in the proposed core micro-architecture. The sparsity-aware processing core is scheduled to be taped out next month. The proposed CIM demonstrations complemented by our previous analysis on analog CIM systems progressed our understanding of this emerging paradigm in pertinence to ML acceleration.

APA, Harvard, Vancouver, ISO, and other styles

13

(7027766), Jonathan A. Fine. "Proton to proteome, a multi-scale investigation of drug discovery." Thesis, 2020.

Find full text

Abstract:

Chemical science spans multiple scales, from a single proton to the collection of proteins that make up a proteome. Throughout my graduate research career, I have developed statistical and machine learning models to better understand chemistry at these different scales, including predicting molecular properties of molecules in analytical and synthetic chemistry to integrating experiments with chemo-proteomic based machine models for drug design. Starting with the proteome, I will discuss repurposing compounds for mental health indications and visualizing the relationships between these disorders. Moving to the cellular level, I will introduce the use of the negative binomial distribution to find biomarkers collected using MS/MS and machine learning models (ML) used to select potent, non-toxic, small molecules for the treatment of castration--resistant prostate cancer (CRPC). For the protein scale, I will introduce CANDOCK, a docking method to rapidly and accurately dock small molecules, an algorithm which was used to create the ML model for CRPC. Next, I will showcase a deep learning model to determine small-molecule functional groups using FTIR and MS spectra. This will be followed by a similar approach used to identify if a small molecule will undergo a diagnostic reaction using mass spectrometry using a chemically interpretable graph-based machine learning method. Finally, I will examine chemistry at the proton level and how quantum mechanics combined with machine learning can be used to understand chemical reactions. I believe that chemical machine learning models have the potential to accelerate several aspects of drug discovery including discovery, process, and analytical chemistry.

APA, Harvard, Vancouver, ISO, and other styles

14

(8788244), Rohan Kumar Manna. "Leakage Conversion For Training Machine Learning Side Channel Attack Models Faster." Thesis, 2020.

Find full text

Abstract:

Recent improvements in the area of Internet of Things (IoT) has led to extensive utilization of embedded devices and sensors. Hence, along with utilization the need for safety and security of these devices also increases proportionately. In the last two decades, the side-channel attack (SCA) has become a massive threat to the interrelated embedded devices. Moreover, extensive research has led to the development of many different forms of SCA for extracting the secret key by utilizing the various leakage information. Lately, machine learning (ML) based models have been more effective in breaking complex encryption systems than the other types of SCA models. However, these ML or DL models require a lot of data for training that cannot be collected while attacking a device in a real-world situation. Thus, in this thesis, we try to solve this issue by proposing the new technique of leakage conversion. In this technique, we try to convert the high signal to noise ratio (SNR) power traces to low SNR averaged electromagnetic traces. In addition to that, we also show how artificial neural networks (ANN) can learn various non-linear dependencies of features in leakage information, which cannot be done by adaptive digital signal processing (DSP) algorithms. Initially, we successfully convert traces in the time interval of 80 to 200 as the cryptographic operations occur in that time frame. Next, we show the successful conversion of traces lying in any time frame as well as having a random key and plain text values. Finally, to validate our leakage conversion technique and the generated traces we successfully implement correlation electromagnetic analysis (CEMA) with an approximate minimum traces to disclosure (MTD) of 480.

APA, Harvard, Vancouver, ISO, and other styles

15

(9868160), Wan-Eih Huang. "Image Processing, Image Analysis, and Data Science Applied to Problems in Printing and Semantic Understanding of Images Containing Fashion Items." Thesis, 2020.

Find full text

Abstract:

This thesis aims to address problems in printing and semantic understanding of images.

The first one is developing a halftoning algorithm for multilevel output with unequal resolution printing pixels. We proposed a design method and implemented several versions of halftone screens. They all show good visual results in a real, low-cost electrophotographic printer.

The second problem is related to printing quality and self-diagnosis. Firstly, we incorporated logistic regression for classification of visible and invisible bands defects in the detection pipeline. In addition, we also proposed a new cost-function based algorithm with synthetic missing bands to estimate the repetitive interval of periodic bands for self-diagnosing the failing component. It is much more accurate than the previous method. Second, we addressed this problem with acoustic signals. Due to the scarcity of printer sounds, an acoustic signal augmentation method is needed to help a classifier perform better. The key idea is to mimic the situation that occurs when a component begins to fail.

The third problem deals with recommendation systems. We explored the similarity metrics in the loss function for a neural matrix factorization network.

The last problem is about image understanding of fashion items. We proposed a weakly supervised framework that includes mask-guided teacher network training and attention-based transfer learning to mitigate the domain gap in datasets and acquire a new dataset with rich annotations.

APA, Harvard, Vancouver, ISO, and other styles

16

(11181642), Deboleena Roy. "Exploring Methods for Efficient Learning in Neural Networks." Thesis, 2021.

Find full text

Abstract:

In the past fifty years, Deep Neural Networks (DNNs) have evolved greatly from a single perceptron to complex multi-layered networks with non-linear activation functions. Today, they form the backbone of Artificial Intelligence, with a diverse application landscape, such as smart assistants, wearables, targeted marketing, autonomous vehicles, etc. The design of DNNs continues to change, as we push its abilities to perform more human-like tasks at an industrial scale.

Multi-task learning and knowledge sharing are essential to human-like learning. Humans progressively acquire knowledge throughout their life, and they do so by remembering, and modifying prior skills for new tasks. In our first work, we investigate the representations learned by Spiking Neural Networks (SNNs), and how to share this knowledge across tasks. Our prior task was MNIST image generation using a spiking autoencoder. We combined the generative half of the autoencoder with a spiking audio-decoder for our new task, i.e audio-to-image conversion of utterances of digits to their corresponding images. We show that objects of different modalities carrying the same meaning can be mapped into a shared latent space comprised of spatio-temporal spike maps, and one can transfer prior skills, in this case, image generation, from one task to another, in a purely Spiking domain. Next, we propose Tree-CNN, an adaptive hierarchical network structure composed of Deep Convolutional Neural Networks(DCNNs) that can grow and learn as new data becomes available. The network organizes the incrementally available data into feature-driven super-classes and improves upon existing hierarchical CNN models by adding the capability of self-growth.

While the above works focused solely on algorithmic design, the underlying hardware determines the efficiency of model implementation. Currently, neural networks are implemented in CMOS based digital hardware such as GPUs and CPUs. However, the saturating scaling trend of CMOS has garnered great interest in Non-Volatile Memory (NVM) technologies such as Spintronics and RRAM. However, most emerging technologies have inherent reliability issues, such as stochasticity and non-linear device characteristics. Inspired by the recent works in spin-based stochastic neurons, we studied the algorithmic impact of designing a neural network using stochastic activations. We trained VGG-like networks on CIFAR-10/100 with 4 different binary activations and analyzed the trade-off between deterministic and stochastic activations.

NVM-based crossbars further promise fast and energy-efficient in-situ matrix-vector multiplications (MVM). However, the analog nature of computing in these NVM crossbars introduces approximations in the MVM operations, resulting in deviations from ideal output values. We first studied the impact of these non-idealities on the performance of vanilla DNNs under adversarial circumstances, and we observed that the non-ideal behavior interferes with the computation of the exact gradient of the model, which is required for adversarial image generation. In a non-adaptive attack, where the attacker is unaware of the analog hardware, analog computing offered varying degree of intrinsic robustness under all attack scenarios - Transfer, Black Box, and White Box attacks. We also demonstrated ``Hardware-in-Loop" adaptive attacks that circumvent this robustness by utilizing the knowledge of the NVM model.

Next, we explored the design of robust DNNs through the amalgamation of adversarial training and the intrinsic robustness offered by NVM crossbar based analog hardware. We studied the noise stability of such networks on unperturbed inputs and observed that internal activations of adversarially trained networks have lower Signal-to-Noise Ratio (SNR), and are sensitive to noise than vanilla networks. As a result, they suffer significantly higher performance degradation due to the non-ideal computations, on an average 2x accuracy drop. On the other hand, for adversarial images, the same networks displayed a 5-10% gain in robust accuracy due to the underlying NVM crossbar when the attack epsilon (the degree of input perturbations) was greater than the epsilon of the adversarial training. Our results indicate that implementing adversarially trained networks on analog hardware requires careful calibration between hardware non-idealities and training epsilon to achieve optimum robustness and performance.

APA, Harvard, Vancouver, ISO, and other styles

17

(8067608), Zhi Li. "COPING WITH LIMITED DATA: MACHINE-LEARNING-BASED IMAGE UNDERSTANDING APPLICATIONS TO FASHION AND INKJET IMAGERY." Thesis, 2019.

Find full text

Abstract:

Machine learning has been revolutionizing our approach to image understanding problems. However, due to the unique nature of the problem, finding suitable data or learning from limited data properly is a constant challenge. In this work, we focus on building machine learning pipelines for fashion and inkjet image analysis with limited data.

We first look into the dire issue of missing and incorrect information on online fashion marketplace. Unlike professional online fashion retailers, sellers on P2P marketplaces tend not to provide correct color categorical information, which is pivotal for fashion shopping. Therefore, to assist users to provide correct color information, we aim to build an image understanding pipeline that can extract garment region in the fashion image and match the color of the fashion item to a pre-defined color categories on the fashion marketplace. To cope with the challenges of lack of suitable data, we propose an autonomous garment color extraction system that uses both clustering and semantic segmentation algorithm to extract the identify fashion garments in the image. In addition, a psychophysical experiment is designed to collect human subjects' color naming schema, and a random forest classifier is trained to given close prediction of color label for the fashion item. Our system is able to perform pixel level segmentation on fashion product portraits and parse human body parts and various fashion categories with human presence.

We also develop an inkjet printing analysis pipeline using pre-trained neural network. Our pipeline is able to learn to perceive print quality, namely high frequency noise level, of the test targets, without intense training. Our research also suggests that in spite of being trained on large scale dataset for object recognition, features generated from neural networks reacts to textural component of the image without any localized features. In addition, we expand our pipeline to printer forensics, and the pipeline is able to identify the printer model by examining the inkjet dot pattern at a microscopic level. Overall, the data-driven computer vision approach presents great value and potential to improve future inkjet imaging technology, even when the data source is limited.

APA, Harvard, Vancouver, ISO, and other styles

18

(8802593), Eric D. Katz. "Differentiating Users Based on Changes in the Underlying Block Space of Their Smartphones." Thesis, 2020.

Find full text

Abstract:

With the growing popularity of using smartphones in business environments, it is increasingly likely that phones will be the target of attacks and sources of evidence in cyber forensic investigations. It will often be important to identify who was using the phone at the time an incident occurred. This can be very difficult as phones are easily misplaced, borrowed, or stolen. Previous research has attempted to find ways to identify computer users based on behavioral analysis. Current research into user profiling requires highly invasive examinations of potentially sensitive user data that the user might not be comfortable with people inspecting or could be against company policy to store. This study developed user profiles based on changes in a mobile phone's underlying block structure. By examining where and when changes occur, a user profile can be developed that is comparable to more traditional intrusion detection models, but without the need to use invasive data sets. These profiles can then be used to determine user masquerading efforts or detect when a compromise has occurred. This study included 35 participants that used Samsung Galaxy S3s for three months. The results of the study show that this method has a high accuracy of classifying a phone's actual sessions correctly when using 2-class models. Results from the 1-class models were not as accurate, but the Sigmoid SVM was able to correctly classify actual user sessions from attack sessions.

APA, Harvard, Vancouver, ISO, and other styles

19

(7011098), Mian Yang. "Quantitative-Scientific Company and Product Scorecard Considerations and Modeling." Thesis, 2019.

Find full text

Abstract:

FDA has long served as the front safeguard to the U.S. citizen public health, is also perceived as one of the world-leading drug regulators. Despite the tremendous efforts and progress have been made to promote the public health, FDA was criticized for putting the agency’s trust icon at stake and was questioned of its ability to serve the agency’s ultimate mission to protect the public. In the wake of the arousing concerns, FDA sought the transformation the oversight model of the medicinal products. One of the actions is to launch quality metrics program. However, this program has been unanimously opposed by the industry. Instead of the current conventional approach, which is constrained by the high dependence on industry cooperation, we try to explore

the measurement of company and product quality risk with public domain data, try to help in visualizing quality and risk. To that end, we develop conceptual frameworks for both company and product quality, examine some of the factors (education, local authority intensity, historical inspection results, physiochemical, physiological, formulation factors, etc.), further developed a warning letter and product recall prediction model with machine learning method referenced to the data analysis outcome.

APA, Harvard, Vancouver, ISO, and other styles

20

(6634508), Amruthavarshini Talikoti. "ESTIMATING PHENYLALANINE OF COMMERCIAL FOODS : A COMPARISON BETWEEN A MATHEMATICAL APPROACH AND A MACHINE LEARNING APPROACH." Thesis, 2019.

Find full text

Abstract:

Phenylketonuria (PKU) is an inherited metabolic disorder affecting 1 in every 10,000 to 15,000 newborns in the United States every year. Caused by a genetic mutation, PKU results in an excessive build up of the amino acid Phenylalanine (Phe) in the body leading to symptoms including but not limited to intellectual disability, hyperactivity, psychiatric disorders and seizures. Most PKU patients must follow a strict diet limited in Phe. The aim of this research study is to formulate, implement and compare techniques for Phe estimation in commercial foods using the information on the food label (Nutritional Fact Label and ordered ingredient list). Ideally, the techniques should be both accurate and amenable to a user friendly implementation as a Phe calculator that would aid PKU patients monitor their dietary Phe intake.

The first approach to solve the above problem is a mathematical one that comprises three steps. The three steps were separately proposed as methods by Jieun Kim in her dissertation. It was assumed that the third method, which is more computationally expensive, was the most accurate one. However, by performing the three methods subsequently in three different steps and combining the results, we actually obtained better results than by merely using the third method.

The first step makes use of the protein content in the foods and Phe:protein multipliers. The second step enumerates all the ingredients in the food and uses the minimum and maximum Phe:protein multipliers of the ingredients along with the protein content. The third step lists the ingredients in decreasing order of their weights, which gives rise to inequality constraints. These constraints hold assuming that there is no loss in the preparation process. The inequality constraints are optimized numerically in two phases. The first involves nutrient content estimation by approximating the ingredient amounts. The second phase is a refinement of the above estimates using the Simplex algorithm. The final Phe range is obtained by performing an interval intersection of the results of the three steps. We implemented all three steps as web applications. Our proposed three-step method yields a high accuracy of Phe estimation (error <= +/- 13.04mg Phe per serving for 90% of foods).

The above mathematical procedure is contrasted against a machine learning approach that uses the data in an existing database as training data to infer the Phe in any given food. Specifically, we use the K-Nearest Neighbors (K-NN) classification method using a feature vector containing the (rounded) nutrient data. In other words, the Phe content of the test food is a weighted average of the Phe values of the neighbors closest to it using the nutrient values as attributes. A four-fold cross validation is carried out to determine the hyper-parameters and the training is performed using the United States Department of Agriculture (USDA) food nutrient database. Our tests indicate that this approach is not very accurate for general foods (error <= +/- 50mg Phe per 100g in about 38% of the foods tested). However, for low-protein foods which are typically consumed by PKU patients, the accuracy increases significantly (error <= +/- 50mg Phe per 100g in over 77% foods).

The machine learning approach is more user-friendly than the mathematical approach. It is convenient, fast and easy to use as it takes into account just the nutrient information. In contrast, the mathematical method additionally takes as input a detailed ingredient list, which is cumbersome to be located in a food database and entered as input. However, the Mathematical method has the added advantage of providing error bounds for the Phe estimate. It is also more accurate than the ML method. This may be due to the fact that for the ML method, the nutrition facts alone are not sufficient to estimate Phe and that additional information like the ingredients list is required.

APA, Harvard, Vancouver, ISO, and other styles

21

(11167785), Nicolae Christophe Iovanac. "GENERATIVE, PREDICTIVE, AND REACTIVE MODELS FOR DATA SCARCE PROBLEMS IN CHEMICAL ENGINEERING." Thesis, 2021.

Find full text

Abstract:

Data scarcity is intrinsic to many problems in chemical engineering due to physical constraints or cost. This challenge is acute in chemical and materials design applications, where a lack of data is the norm when trying to develop something new for an emerging application. Addressing novel chemical design under these scarcity constraints takes one of two routes: the traditional forward approach, where properties are predicted based on chemical structure, and the recent inverse approach, where structures are predicted based on required properties. Statistical methods such as machine learning (ML) could greatly accelerate chemical design under both frameworks; however, in contrast to the modeling of continuous data types, molecular prediction has many unique obstacles (e.g., spatial and causal relationships, featurization difficulties) that require further ML methods development. Despite these challenges, this work demonstrates how transfer learning and active learning strategies can be used to create successful chemical ML models in data scarce situations.

Transfer learning is a domain of machine learning under which information learned in solving one task is transferred to help in another, more difficult task. Consider the case of a forward design problem involving the search for a molecule with a particular property target with limited existing data, a situation not typically amenable to ML. In these situations, there are often correlated properties that are computationally accessible. As all chemical properties are fundamentally tied to the underlying chemical topology, and because related properties arise due to related moieties, the information contained in the correlated property can be leveraged during model training to help improve the prediction of the data scarce property. Transfer learning is thus a favorable strategy for facilitating high throughput characterization of low-data design spaces.

Generative chemical models invert the structure-function paradigm, and instead directly suggest new chemical structures that should display the desired application properties. This inversion process is fraught with difficulties but can be improved by training these models with strategically selected chemical information. Structural information contained within this chemical property data is thus transferred to support the generation of new, feasible compounds. Moreover, transfer learning approach helps ensure that the proposed structures exhibit the specified property targets. Recent extensions also utilize thermodynamic reaction data to help promote the synthesizability of suggested compounds. These transfer learning strategies are well-suited for explorative scenarios where the property values being sought are well outside the range of available training data.

There are situations where property data is so limited that obtaining additional training data is unavoidable. By improving both the predictive and generative qualities of chemical ML models, a fully closed-loop computational search can be conducted using active learning. New molecules in underrepresented property spaces may be iteratively generated by the network, characterized by the network, and used for retraining the network. This allows the model to gradually learn the unknown chemistries required to explore the target regions of chemical space by actively suggesting the new training data it needs. By utilizing active learning, the create-test-refine pathway can be addressed purely in silico. This approach is particularly suitable for multi-target chemical design, where the high dimensionality of the desired property targets exacerbates data scarcity concerns.

The techniques presented herein can be used to improve both predictive and generative performance of chemical ML models. Transfer learning is demonstrated as a powerful technique for improving the predictive performance of chemical models in situations where a correlated property can be leveraged alongside scarce experimental or computational properties. Inverse design may also be facilitated through the use of transfer learning, where property values can be connected with stable structural features to generate new compounds with targeted properties beyond those observed in the training data. Thus, when the necessary chemical structures are not known, generative networks can directly propose them based on function-structure relationships learned from domain data, and this domain data can even be generated and characterized by the model itself for closed-loop chemical searches in an active learning framework. With recent extensions, these models are compelling techniques for looking at chemical reactions and other data types beyond the individual molecule. Furthermore, the approaches are not limited by choice of model architecture or chemical representation and are expected to be helpful in a variety of data scarce chemical applications.

APA, Harvard, Vancouver, ISO, and other styles

22

(5930033), Curtis P. Martin. "RATIONAL DESIGN OF TYPE II KINASE INHIBITORS VIA NOVEL MULTISCALE VIRTUAL SCREENING APPROACH." Thesis, 2019.

Find full text

Abstract:

At present, the combination of high drug development costs and external pressure to lower consumer prices is forcing the pharmaceutical industry to innovate in ways unlike ever before. One of the main drivers of increased productivity in research and development recently has been the application of computational methods to the drug discovery process. While this investment has generated promising insights in many cases, there is still much progress to be made.

There currently exists a dichotomy in the types of algorithms employed which are roughly defined by the extent to which they compromise predictive accuracy for computational efficiency, and vice versa. Many computational drug discovery algorithms exist which yield commendable predictive power but are typically associated with overwhelming resource costs. High-throughput methods are also available, but often suffer from disappointing and inconsistent performance.

In the world of kinase inhibitor design, which often takes advantage of such computational tools, small molecules tend to have myriad side effects. These are usually caused by off-target binding, especially with other kinases (given the large size of the enzyme family and overall structural conservation), and so inhibitors with tunable selectivity are generally desirable. This issue is compounded when considering therapeutically relevant targets like Abelson Protein Tyrosine Kinase (Abl) and Lymphocyte Specific Protein Tyrosine Kinase (Lck) which have opposing effects in many cancers.

This work attempts to solve both of these problems by creating a methodology which incorporates high-throughput computational drug discovery methods, modern machine learning techniques, and novel protein-ligand binding descriptors based on backbone hydrogen bond (dehydron) wrapping, chosen because of their potential in differentiating between kinases. Using this approach, a procedure was developed to quickly screen focused chemical libraries (in order to narrow the domain of applicability and keep medicinal chemistry at the forefront of development) for detection of selective kinase inhibitors. In particular, five pharmacologically relevant kinases were investigated to provide a proof of concept, including those listed above.

Ultimately, this work shows that dehydron wrapping indeed has predictive value, though it's likely hindered by common and current issues derived from noisy training data, imperfect feature selection algorithms, and simplifying assumptions made by high-throughput algorithms used for structural determination. It also shows that the procedure's predictive value varies depending on the target, leading to the conclusion that the utility of dehydron wrapping for drug design is not necessarily universal, as originally thought. However, for those targets which are amenable to the concept, there are two major benefits: relatively few features are required to produce modest results; and those structural features chosen are easily interpretable and can thereby improve the overall design process by pointing out regions to optimize within any given lead. Of the five kinases explored, Src and Lck are shown in this work to fit particularly well with the general hypothesis; given their importance in treating cancer and evading off-target related side effects, the developed methodology now has the potential to play a major role in the development of drug candidates which specifically inhibit and avoid these kinases.

APA, Harvard, Vancouver, ISO, and other styles

23

(7036478), Aiyshwariya Paulvannan Kanmani. "A data-centric framework for assessing environmental sustainability." Thesis, 2019.

Find full text

Abstract:

Necessity to sustain resources has risen in recent years with significant number of people affected by lack of access to essential resources. Framing policies that support environmental sustainability is necessary for addressing the issue. Effective policies necessitate access to a framework which assesses and keeps track of sustainability. Conventional frameworks that support such policy-making involve ranking of countries based on a weighted sum of several environmental performance metrics. However, the selection and weighing of metrics is often biased. This study proposes a new framework to assess environmental sustainability of countries via leveraging unsupervised learning. Specifically, this framework harnesses a clustering technique and tracks progressions in terms of shifts within clusters over time. It is observed that using the proposed framework, countries can identify specific ways to improve their progress towards environmental sustainability.

APA, Harvard, Vancouver, ISO, and other styles

24

(6861362), Yang Yan. "Image-Based Non-Contact Conductivity Prediction for Inkjet Printed Electrodes and Follow-Up Work of Toner Usage Prediction for Laser Electro-Phorographic Printers." Thesis, 2019.

Find full text

Abstract:

This thesis includes two parts. The main part is on the topic of conductivity prediction for Inkjet printed silver electrodes. The second part is about the follow-up work of toner usage prediction of laser electro-photographic printers.

For conductivity prediction of Inkjet printed silver electrodes part, the brief introduction is described below. Recently, electronic devices made with Inkjet printing technique and flexible thin films have attracted great attention due to their potential applications in sensor manufacturing. This imaging system has become a great tool to monitor the quality of Inkjet printed electrodes due to the fact that most thickness or resistance measuring devices can destroy the surface of a printed electrode or even whole electrode. Thus, a non-contact image-based approach to estimate sheet resistance of Inkjet printed electrodes is developed.

The approach has two stages. Firstly, strip-shaped electrodes are systematically printed with various printing parameters. The sheet resistance measurement data as

well as images of the electrodes are acquired. Then, based on the real experimental data, the fitting model is constructed and further used in predicting the sheet

resistance of the Inkjet printed silver electrodes.

For toner usage prediction part, the introduction is described below. With the widespread use of laser electro-photographic printers in both industry and households fields, estimation of toner usage has great significance to ensuring the full utilization of each cartridge. The follow-up work is focused on testing and improving feasibility, reliability, and adaptability of the Black Box Model (BBM) based two-stage strategy in estimating the toner usage. Comparing with previous methods, the training process for the firrst stage requires less time and disk storage, all while maintaining high accuracy. For the second stage, experiments are performed on various models of printers, with cyan(C), magenta(M), yellow(Y), and black(K) color cartridges.

APA, Harvard, Vancouver, ISO, and other styles

25

(10695907), Wo Jae Lee. "AI-DRIVEN PREDICTIVE WELLNESS OF MECHANICAL SYSTEMS: ASSESSMENT OF TECHNICAL, ENVIRONMENTAL, AND ECONOMIC PERFORMANCE." Thesis, 2021.

Find full text

Abstract:

One way to reduce the lifecycle cost and environmental impact of a product in a circular economy is to extend its lifespan by either creating longer-lasting products or managing the product properly during its use stage. Life extension of a product is envisioned to help better utilize raw materials efficiently and slow the rate of resource depletion. In the case of manufacturing equipment (e.g., an electric motor on a machine tool), securing reliable service life as well as the life extension are important for consistent production and operational excellence in a factory. However, manufacturing equipment is often utilized without a planned maintenance approach. Such a strategy frequently results in unplanned downtime, owing to unexpected failures. Scheduled maintenance replaces components frequently to avoid unexpected equipment stoppages, but increases the time associated with machine non-operation and maintenance cost.

Recently, the emergence of Industry 4.0 and smart systems is leading to increasing attention to predictive maintenance (PdM) strategies that can decrease the cost of downtime and increase the availability (utilization rate) of manufacturing equipment. PdM also has the potential to foster sustainable practices in manufacturing by maximizing the useful lives of components. In addition, advances in sensor technology (e.g., lower fabrication cost) enable greater use of sensors in a factory, which in turn is producing greater and more diverse sets of data. Widespread use of wireless sensor networks (WSNs) and plug-and-play interfaces for the data collection on product/equipment states are allowing predictive maintenance on a much greater scale. Through advances in computing, big data analysis is faster/improved and has allowed maintenance to transition from run-to-failure to statistical inference-based or machine learning prediction methods.

Moreover, maintenance practice in a factory is evolving from equipment “health management” to equipment “wellness” by establishing an integrated and collaborative manufacturing system that responds in real-time to changing conditions in a factory. The equipment wellness is an active process of becoming aware of the health condition and of making choices that achieve the full potential of the equipment. In order to enable this, a large amount of machine condition data obtained from sensors needs to be analyzed to diagnose the current health condition and predict future behavior (e.g., remaining useful life). If a fault is detected during this diagnosis, a root cause of a fault must be identified to extend equipment life and prevent problem reoccurrence.

However, it is challenging to build a model capturing a relationship between multi-sensor signals and mechanical failures, considering the dynamic manufacturing environment and the complex mechanical system in equipment. Another key challenge is to obtain usable machine condition data to validate a method.

A goal of the proposed work is to develop a systematic tool for maintenance in manufacturing plants using emerging technologies (e.g., AI, Smart Sensor, and IoT). The proposed method will facilitate decision-making that supports equipment maintenance by rapidly detecting a worn component and estimating remaining useful life. In order to diagnose and prognose a health condition of equipment, several data-driven models that describe the relationships between proxy measures (i.e., sensor signals) and machine health conditions are developed and validated through the experiment for several different manufacturing-oriented cases (e.g., cutting tool, gear, and bearing). To enhance the robustness and the prediction capability of the data-driven models, signal processing is conducted to preprocess the raw signals using domain knowledge. Through this process, useful features from the large dataset are extracted and selected, thus increasing computational efficiency in model training. To make a decision using the processed signals, a customized deep learning architecture for each case is designed to effectively and efficiently learn the relationship between the processed signals and the model’s outputs (e.g., health indicators). Ultimately, the method developed through this research helps to avoid catastrophic mechanical failures, products with unacceptable quality, defective products in the manufacturing process as well as to extend equipment service life.

To summarize, in this dissertation, the assessment of technical, environmental and economic performance of the AI-driven method for the wellness of mechanical systems is conducted. The proposed methods are applied to (1) quantify the level of tool wear in a machining process, (2) detect different faults from a power transmission mini-motor testbed (CNN), (3) detect a fault in a motor operated under various rotation speeds, and (4) to predict the time to failure of rotating machinery. Also, the effectiveness of maintenance in the use stage is examined from an environmental and economic perspective using a power efficiency loss as a metric for decision making between repair and replacement.

APA, Harvard, Vancouver, ISO, and other styles

Dissertations / Theses on the topic 'Machine learning not elsewhere classified'

Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles