A Survey to Time Series Anomaly Detection in Different Data Structures

Communications in computer and information science, Journal Year: 2025, Volume and Issue: unknown, P. 206 - 217

Published: Jan. 1, 2025

Language: Английский

---

Artificial Immune System for Fault Detection and Classification of Semiconductor Equipment

Electronics, Journal Year: 2021, Volume and Issue: 10(8), P. 944 - 944

Published: April 15, 2021

Semiconductor manufacturing comprises hundreds of consecutive unit processes. A single misprocess could jeopardize the whole process. In current environments, data monitoring equipment condition, wafer metrology, and inspection, etc., are used to probe any anomaly during process that affect final chip performance quality. The purpose investigation is fault detection classification (FDC). Various methods, such as statistical or mining methods with machine learning algorithms, have been employed for FDC. this paper, we propose an artificial immune system (AIS), which a biologically inspired computing algorithm, FDC regarding semiconductor equipment. Process shifts caused by parts modules aging over time main processes failure cause. We employ state variable identification (SVID) data, contain operating optical emission spectroscopy (OES) represent plasma information obtained from faulty scenario intentional modification gas flow rate in fabrication achieved modeling prediction accuracy 94.69% selected SVID OES 93.68% alone. To conclude, possibility using AIS field decision making proposed.

Language: Английский

---

Enhancing Sound-Based Anomaly Detection Using Deep Denoising Autoencoder

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 84323 - 84332

Published: Jan. 1, 2024

Language: Английский

---

Humor Detection System for MuSE 2023: Contextual Modeling, Pesudo Labelling, and Post-smoothing

Published: Oct. 20, 2023

Humor detection has emerged as an active research area within the field of artificial intelligence. Over past few decades, it made remarkable progress with development deep learning. This paper introduces a novel framework aimed at enhancing model's understanding humorous expressions. Specifically, we consider impact correspondence between labels and features. In order to achieve more effective models limited training samples, employ widely utilized semi-supervised learning technique called pseudo labeling. Furthermore, use post-smoothing strategy eliminate abnormally high predictions. At same time, in alleviate over-fitting phenomenon model on validation set, created 10 different random subsets then aggregating their prediction. To verify effectiveness our strategy, evaluate its performance Cross-Cultural Humour sub-challenge MuSe 2023. Experimental results demonstrate that system achieves AUC score 0.9112, surpassing baseline by substantial margin.

Language: Английский

---

LSTM-based framework with metaheuristic optimizer for manufacturing process monitoring

Alexandria Engineering Journal, Journal Year: 2023, Volume and Issue: 83, P. 43 - 52

Published: Oct. 25, 2023

Quick process shift detection and lower out-of-control run length are essential for monitoring the production process, especially in modern smart manufacturing. Specifically, is one of most critical performance measures to evaluate manufacturing (MPM) model. The sooner detected, better model is. However, developing a which can provide quick various data dimensions volumes challenging. In this research, single (1_LSTM) stacked (S_LSTM) long-short-term memory (LSTM) based models with metaheuristic optimizer were proposed detect shifts quickly domain. Based on literature, three methods: Clustering-based organism search (CSOS), Particle Swarm Optimization (PSO), Simulated Annealing (SA) that suitable high-dimensional optimization utilized method optimize weights LSTM-based network. evaluated average (ARL1) against benchmark methods synthesized multivariate normal real-world datasets. Also, performances CSOS, PSO, SA compared. results show CSOS_S_LSTM outperforms other ARL1. result also confirmed effectiveness applicability problems. experimental showed response time be improved by 33.19% 38.77% using 1_LSTM CSOS metaheuristics models, respectively.

Language: Английский

---

MVOPFAD: Multiview Online Passenger Flow Anomaly Detection

IEEE Sensors Journal, Journal Year: 2024, Volume and Issue: 24(9), P. 14668 - 14681

Published: March 18, 2024

Prompt and accurate identification of anomalies in passenger flow within metro systems is crucial for safety, security, operational efficiency. However, traditional anomaly detection methods often struggle to achieve high accuracy low latency when constrained by limited labeled data online applications. This study presents a straightforward yet effective framework, termed multiview (MVOPFAD), address these difficulties data-driven manner. Specifically, reduce the computational burden meet requirements, we particularly propose an elastic extreme studentized deviate (EESD) model accounting characteristic abnormal flow. Concurrently, improved shifted wavelet tree (ISWT) employed effectively capture various features. It joined implementation ensemble learning techniques EESD further enhance robustness our model. To evaluate performance proposed conducted comprehensive series experiments utilizing collected from automated fare collection (AFC) system integrated into Beijing Metro network. Our MVOPFAD demonstrates significant superiority over other three types across all evaluation metrics. In particular, it yields 15.49% increase precision 9.71% rise $F2$ -score compared second-best detecting outbound anomalies. Additionally, incurs lower cost than deep models machine models. The experimental results strongly suggest feasibility implementation, thereby demonstrating practicality effectiveness

Language: Английский

---

Model-free detection of unique events in time series

Scientific Reports, Journal Year: 2022, Volume and Issue: 12(1)

Published: Jan. 7, 2022

Recognition of anomalous events is a challenging but critical task in many scientific and industrial fields, especially when the properties anomalies are unknown. In this paper, we introduce new anomaly concept called "unicorn" or unique event present new, model-free, unsupervised detection algorithm to detect unicorns. The key component Temporal Outlier Factor (TOF) measure uniqueness continuous data sets from dynamic systems. differs significantly traditional outliers aspects: while repetitive no longer events, not necessarily an outlier; it does fall out distribution normal activity. performance our was examined recognizing on different types simulated with compared Local (LOF) discord discovery algorithms. TOF had superior LOF algorithms even also recognized that those did not. benefits unicorn method were illustrated by example very fields. Our successfully cases where they already known such as gravitational waves binary black hole merger LIGO detector signs respiratory failure ECG series. Furthermore, found LIBOR set last 30 years.

Language: Английский

---

An Enterprise Time Series Forecasting System for Cloud Applications Using Transfer Learning

Sensors, Journal Year: 2021, Volume and Issue: 21(5), P. 1590 - 1590

Published: Feb. 25, 2021

The main purpose of an application performance monitoring/management (APM) software is to ensure the highest availability, efficiency and security applications. An APM accomplishes goals through automation, measurements, analysis diagnostics. Gartner specifies three crucial capabilities softwares. first end-user experience monitoring for revealing interactions users with infrastructure components. second discovery, diagnostics tracing. third key component machine learning (ML) artificial intelligence (AI) powered data analytics predictions, anomaly detection, event correlations root cause analysis. Time series metrics, logs traces are pillars observability valuable source information IT operations. Accurate, scalable robust time forecasting detection requested analytics. Approaches based on neural networks (NN) deep gain increasing popularity due their flexibility ability tackle complex nonlinear problems. However, some disadvantages NN-based models distributed cloud applications mitigate expectations require specific approaches. We demonstrate how NN-models, pretrained a global database, can be applied customer using transfer learning. In general, NN-models adequately operate only stationary series. Application nonstationary requires multilayer processing including hypothesis testing categorization, category transformations into data, backward transformations. present mathematical background this approach discuss experimental results implementation Wavefront by VMware (an software) while real environments.

Language: Английский

---

Incident Management for Explainable and Automated Root Cause Analysis in Cloud Data Centers    

JUCS - Journal of Universal Computer Science, Journal Year: 2021, Volume and Issue: 27(11), P. 1152 - 1173

Published: Nov. 28, 2021

Effective root cause analysis (RCA) of performance issues in modern cloud environ- ments remains a hard problem. Traditional RCA tracks complex by their signatures known as problem incidents. Common approaches to incident discovery rely mainly on expertise users who define environment-specific set alerts and >target detection problems through occurrence the monitoring system. Adequately modeling all possible patterns for nowadays extremely sophisticated data center applications is very task. It may result alert/event storms including large numbers non-indicative precautions. Thus, crucial task incident-based reduction redundant recommendations prioritizing those events subject importance/impact criteria or deriving meaningful groupings into separable situations. In this paper, we consider automation based rule induction algorithms that retrieve conditions directly from datasets without consuming sys- tem events. Rule-learning are flexible powerful many regression classification problems, with high-level explainability. Since annotated labeled sets mostly unavailable area technology, discuss self-labelling principles which allow transforming originally unsupervised learning tasks further application methods detection.

Language: Английский

---

Machine Learning-Based Time-Series Data Analysis in Edge-Cloud-Assisted Oil Industrial IoT System

Mobile Information Systems, Journal Year: 2022, Volume and Issue: 2022, P. 1 - 11

Published: Feb. 17, 2022

With the rapid development of Industrial Internet Things (IIoT) and edge computing techniques, in situ intelligent sensors are continuously generating increasing vast amounts time-series data. In many industrial applications, particularly highly distributed systems positioned remote areas, repeated transmission large raw data onto server is not possible. This poses a significant challenge to timely processing these IIoT. Analyzing all remotely cloud impractical has very low efficiency owing network latency limited resources. Failure detecting abnormal may result major production safety problems. Therefore, businesses moving machine learning capabilities enable real-time actions field. this study, we present machine-learning-based edge-cloud framework solve problem. First, robust random cut forest isolation algorithms employed at gateway collected for detection anomalously changing Subsequently, preprocessed transmitted services trend prediction missing completion using long short-term memory recurrent neural method feed conjunction with original sequence historical combined first-order forward difference The experimental results show that edge-cloud-assisted oil IIoT system can improve substantially accuracy analyses.

Language: Английский

---