Graphical tools for routine assessment of industrial alarm systems

Kondaveeti, Sandeep R.; Izadi, Iman; Shah, Sirish L.; Black, Tim; Chen, Tongwen

doi:10.1016/j.compchemeng.2012.06.042

Cited by 59 publications

(30 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In particular, the calculated Jaccard or Sorgenfrei coefficient could be very low even for the case that two alarm sequences are indeed correlated. This issue had been noticed previously and two solutions were proposed (Kondaveeti et al, 2012;Yang et al, 2012). One solution is to pad extra 1's at both sides of each 1 in x a (t) (Kondaveeti et al, 2012).…”

Section: A N U S C R I P Tmentioning

confidence: 92%

“…This issue had been noticed previously and two solutions were proposed (Kondaveeti et al, 2012;Yang et al, 2012). One solution is to pad extra 1's at both sides of each 1 in x a (t) (Kondaveeti et al, 2012). That is, if x a (t) = 1, then extra 1's…”

Section: A N U S C R I P Tmentioning

confidence: 93%

“…The presence of occurrence delays has significant effects on the existing methods to detect correlated alarms, which are based on similarity measures of binary sequences such as Jaccard and Sorgenfrei coefficients (Kondaveeti et al, 2012;Yang, Wang & Chen, 2013). The definitions of Jaccard and Sorgenfrei coefficients are…”

Section: A N U S C R I P Tmentioning

confidence: 99%

“…Yang, Wang & Chen (2013) compared different similarity coefficients for binary-valued alarm signals, and detected correlated alarms based on the Sorgenfrei coefficient and the distribution of correlation delays. This paper is a continuing study of our previous work (Yang, Wang & Chen, 2013), and is motivated by two drawbacks revealed in our industrial experience in the application of the method proposed in Yang, Wang & Chen (2013) to detect correlated alarms: (i) The estimated Sorgenfrei coefficients (as well as Jaccard coefficients in Kondaveeti et al (2012)) sometimes are quite small even for two alarm signals that have been known to be physically related; (ii) the distribution of correlation delays requires a large amount A c c e p t e d M a n u s c r i p t of data samples to get a sufficient number of alarm activations, which sometimes is a demanding requirement in practice. This paper analyzes the causes of the two drawbacks and proposes corresponding solutions.…”

Section: Introductionmentioning

confidence: 99%

“…Noda et al (2011) proposed a scheme called event correlation analysis to quantify the relationship between alarms and operating actions, and to identify correlated alarms and unnecessary operating actions. Kondaveeti et al (2012) grouped similar alarms together based on Jaccard similarity coefficients in an alarm similarity map. Yang et al (2012) transformed binary alarm sequences to continuous-valued pseudo alarm sequences and clustered correlated alarms based on the cross correlation of pseudo alarm sequences.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

A new method to detect and quantify correlated alarms with occurrence delays

Wang

Chen

2015

Computers & Chemical Engineering

Self Cite

View full text Add to dashboard Cite

Section: A N U S C R I P Tmentioning

confidence: 92%

Section: A N U S C R I P Tmentioning

confidence: 93%

Section: A N U S C R I P Tmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A new method to detect and quantify correlated alarms with occurrence delays

Wang

Chen

2015

Computers & Chemical Engineering

Self Cite

View full text Add to dashboard Cite

Understanding the importance of process alarms based on the analysis of deep recurrent neural networks trained for fault isolation

Dörgő

Pigler

Abonyi

2018

Journal of Chemometrics

View full text Add to dashboard Cite

The identification of process faults is a complex and challenging task due to the high amount of alarms and warnings of control systems. To extract information about the relationships between these discrete events, we utilise multitemporal sequences of alarm and warning signals as inputs of a recurrent neural network-based classifier and visualise the network by principal component analysis. The similarity of the events and their applicability in fault isolation can be evaluated based on the linear embedding layer of the network, which maps the input signals into a continuous-valued vector space. The method is demonstrated in a simulated vinyl acetate production technology. The results illustrate that with the application of recurrent neural network-based sequence learning not only accurate fault classification solutions can be developed, but the visualisation of the model can give useful hints for hazard analysis. KEYWORDS deep learning, fault classification, visualisation of discrete eventsNomenclature: RNN, -recurrent neural network; LSTM, -long short-term memory; PCA, -principal component analysis; VAc, -vinyl acetate; y, -type of fault, = { c 1 , … , c n c } , related to the kth sequence of events.;̂k, -predicted class label of the faults; c j , -jth type fault; n c , -number of fault types; s, -state of the technology; pv, -index of the process variable; a, -the related state signals; e, -event; st, -starting time of an event; et, -ending time of an event; k , -kth sequence of states; S, -the set of states; R, -arbitrary temporal predicate between events; E, -equal temporal predicates between events; B, -before temporal predicates between events; D, -during temporal predicates between events; O, -overlap temporal predicates between events; x k , -the numerical representation of k sequence; P(y k = c j |x k ), -the conditional probability of the given fault; h k ,vector of the activities of the LSTM units; n U , -the number of LSTM units in the RNN; T, -length of the input sequence; W s , -weight matrix of the output layer of the network (referring to the applied softmax function); C t k , -cell state of the tth LSTM unit; t k , -forget gate of the tth LSTM unit; W f , -weight matrix of the forget gate; b, -the bias vector of the corresponding neurons; i t k , -input gate of the tth LSTM unit; o t k , -output gate of the tth LSTM unit; W o , -weight matrix of the output gate; oh t k , -one-hot binary vector representation of the tth symbol of the kth sequence; n s , -the number of states in the D temporal database; n R , -number of types of temporal predicates between events; n o , -number of bits in the one-hot binary vector; n e , -dimension of the embedding layer; W, -weight matrix of embedding layer; sim(s i , s j ), -similarity of the alarms; d max , -the maximum Euclidean distance among the rows of the W matrix;

show abstract

Dot Matrix Analysis of Plant Operation Data for Identifying Sequential Alarms Triggered by Single Root Cause

Wang

Noda

2015

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Graphical tools for routine assessment of industrial alarm systems

Cited by 59 publications

References 7 publications

A new method to detect and quantify correlated alarms with occurrence delays

A new method to detect and quantify correlated alarms with occurrence delays

Understanding the importance of process alarms based on the analysis of deep recurrent neural networks trained for fault isolation

Dot Matrix Analysis of Plant Operation Data for Identifying Sequential Alarms Triggered by Single Root Cause

Contact Info

Product

Resources

About