Capturing Causality for Fault Diagnosis Based on Multi-Valued Alarm Series Using Transfer Entropy

Su, Jianjun; Wang, Dezheng; Zhang, Yinong; Yang, Fan; Zhao, Yan; Pang, Xiaolu

doi:10.3390/e19120663

Cited by 19 publications

(6 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The era of big data in process industries is coming, resulting in the increasing complexity of modern industrial systems. Therefore, industrial safety is confronted with intricate challenges within the framework of the extensive deployment of sensors [18]. A reliable fault detection system is urgently required to avoid breakdowns or sudden failures.…”

Section: Related Workmentioning

confidence: 99%

A lightweight complex-domain acoustic feature extraction method for rotating machinery fault detection

Wei,

Ding,

Wang

et al. 2024

Meas. Sci. Technol.

Self Cite

View full text Add to dashboard Cite

In the field of fault diagnosis for factory machinery systems, the development of deep learning methods has been hindered by the challenge of acquiring fault data, highlighting the need to extract noise robust features from limited labeled data. In this paper, a light and efficient complex-domain acoustic feature extraction method, complex-domain principal frequency convolution network (CPFCN) is proposed for fault diagnosis in rotating machinery, which consists of a principal frequency filter (PFF) and stacked convolution network (SCN). The PFF filters out non-principal frequency noise to focus on the predominant frequency. The SCN is designed to effectively extract the amplitude and phase features, which can fully leverage the complex-domain information within the acoustic data. The experimental results show that the proposed CPFCN have 33% increasing in accuracy while 87% reduction in training time and 41% reduction in feature extraction time. Additionally, the proposed framework has improved the accuracy by 59% on the dataset with noise compared to the best-performing method in the experimental study, achieving stronger noise robustness in the case of limited samples.

show abstract

Section: Related Workmentioning

confidence: 99%

A lightweight complex-domain acoustic feature extraction method for rotating machinery fault detection

Wei,

Ding,

Wang

et al. 2024

Meas. Sci. Technol.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Hu et al present an RCA framework including process data and alarm configuration analysis, thus targeting a broader scope than this letter, but do not focus use case-specific adaptability or iterative analysis of evolving alarm data [14]. RCA algorithms often build upon statistically based clustering, e.g., using graph similarities [15], transfer entropy between temporally linked alarms [13], [16], cyclically occurring temporal patterns [17], or distance measures between graphs [18]. All above-mentioned algorithms map a series of logged alarms to highly correlated alarm sets, considered as candidates for causal sequences.…”

Section: Related Workmentioning

confidence: 99%

A Distributed Framework for Knowledge-Driven Root-Cause Analysis on Evolving Alarm Data–An Industrial Case Study

Wilch

Vogel‐Heuser

Mager³

et al. 2023

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

Root-cause Analysis (RCA) of alarms is a wellestablished research area in automated Production Systems (aPS).Many RCA algorithms have been proposed and successfully evaluated and new ones are being developed. Recently, researchers focus on the incorporation of formalized information about the technical process in the analysis to gather further evidence for common root causes. In industrial applications, alarm data are usually preprocessed to accommodate for use case-specific properties and prepare subsequent analysis steps. Consequently, this letter proposes a generalized RCA framework, for which an arbitrary number of preprocessing, data-driven RCA, and postprocessing algorithms can be selected, to support varying use cases. The framework was successfully evaluated in an industrial case study, using 1.8 million alarms recorded over 450 days from an industrial nonwoven production plant and analyzed using formalized information from process documentation and expert interviews. Seven preprocessing algorithms, one data-driven RCA algorithm, and nine postprocessing algorithms typical for continuous and hybrid technical processes were realized in an otherwise entirely use case-agnostic implementation.Index Terms-Software architecture for robotic and automation, Big Data in robotics and automation.

show abstract

“…Accordingly, causality inference, association rule mining, and sequential pattern mining have been exploited to find such directions when detecting correlated alarms. Among these approaches, causality inference detects the causal relations from historical data complemented by process knowledge [12]; commonly used methods include Transfer Entropy [13][14][15], Granger causality [16], and qualitative trend analysis [17]. In [14,15], transfer entropies were exploited and modified to detect the causal relations between alarm signals.…”

Section: Introductionmentioning

confidence: 99%

A First-Out Alarm Detection Method via Association Rule Mining and Correlation Analysis

Li,

Cheng

2023

Entropy

View full text Add to dashboard Cite

Alarm systems are commonly deployed in complex industries to monitor the operation status of the production process in real time. Actual alarm systems generally have alarm overloading problems. One of the major factors leading to excessive alarms is the presence of many correlated or redundant alarms. Analyzing alarm correlations will not only be beneficial to the detection of and reduction in redundant alarm configurations, but also help to track the propagation of abnormalities among alarm variables. As a special problem in correlated alarm detection, the research on first-out alarm detection is very scarce. A first-out alarm is known as the first alarm that occurs in a series of alarms. Detection of first-out alarms aims at identifying the first alarm occurrence from a large number of alarms, thus ignoring the subsequent correlated alarms to effectively reduce the number of alarms and prevent alarm overloading. Accordingly, this paper proposes a new first-out alarm detection method based on association rule mining and correlation analysis. The contributions lie in the following aspects: (1) An association rule mining approach is presented to extract alarm association rules from historical sequences based on the FP-Growth algorithm and J-Measure; (2) a first-out alarm determination strategy is proposed to determine the first-out alarms and subsequent alarms through correlation analysis in the form of a hypothesis test on conditional probability; and (3) first-out rule screening criteria are proposed to judge whether the rules are redundant or not and then consolidated results of first-out rules are obtained. The effectiveness of the proposed method is tested based on the alarm data generated by a public simulation platform.

show abstract

Capturing Causality for Fault Diagnosis Based on Multi-Valued Alarm Series Using Transfer Entropy

Cited by 19 publications

References 31 publications

A lightweight complex-domain acoustic feature extraction method for rotating machinery fault detection

A lightweight complex-domain acoustic feature extraction method for rotating machinery fault detection

A Distributed Framework for Knowledge-Driven Root-Cause Analysis on Evolving Alarm Data–An Industrial Case Study

A First-Out Alarm Detection Method via Association Rule Mining and Correlation Analysis

Contact Info

Product

Resources

About