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

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 multi-temporal sequences of alarm and warning signals as inputs of a recurrent neural network (RNN) 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 RNN 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.

The paper will be published in the Journal of Chemometrics, 2018.

The manuscript and the Python codes can be downloaded from our Github repository.

Learning operation strategies from alarm management systems by temporal pattern mining and deep learning

We introduce a sequence to sequence deep learning algorithm to learn and predict sequences of process alarms and warnings. The proposed recurrent neural network model utilizes an encoder layer of Long Short-Term Memory (LSTM) units to map the input sequence of discrete events into a vector of fixed dimensionality, and a decoder LSTM layer to form a prediction of the sequence of future events. We demonstrate that the information extracted by this model from alarm log databases can be used to suppress alarms with low information content which reduces the operator workload.

To generate easily reproducible results and stimulate the development of alarm management algorithms we define an alarm management benchmark problem based on the simulator of a vinyl acetate production technology. The results confirm that sequence to sequence learning is a useful tool in alarm rationalization and, in more general, for process engineers interested in predicting the occurrence of discrete events.

The paper will be published in the Proceedings of the 28th European Symposium on Computer Aided Process Engineering (ESCAPE), 2018.

The article will be downloadable from the website of the journal or from my ResearchGate profile after the appearance of the onlinefirst version.

Towards Operator 4.0, increasing production efficiency and reducing operator workload by process mining of alarm data

We propose a methodology to extract temporal patterns of alarm sequences and operator actions from the log files of alarm management systems. Firstly, the algorithm identifies time-segments that informative from the viewpoint of operator interventions. These segments include series of alarms that initialise operator actions, sets of operator actions, and a period that potentially covers the effects of the corrective actions of the operators. In the second step of the methodology, we discover the sets of the operator actions frequently applied in the same situations. For this purpose, we utilise the FP-Growth algorithm, which is one of the fastest tools of frequent itemset mining and generates well-structured action trees that are not only suitable for the visualisation of the interventions but lend themselves to build association rules that could be directly applied in decision support systems. Finally, we apply multi-temporal sequence mining to reveal what alarms led to the sets of operator actions and what were the effects of these interventions. We illustrate the applicability of the methodology by presenting illustrative results connected to the analysis of the delayed coker plant in the MOL Group Danube Refinery.

The paper will be published in the Chemical Engineering Transactions, 2018.

The article will be downloadable from the website of the journal or from my ResearchGate profile after the appearance of the onlinefirst version.

Sequence Mining based Alarm Suppression

To provide more insight into the process dynamics and represent the temporal relationships among faults, control actions and process variables we propose of a multi-temporal sequence mining based algorithm. The methodology starts with the generation of frequent temporal patterns of the alarm signals. We transformed the multi-temporal sequences into Bayes classifiers. The obtained association rules can be used to define alarm suppression rules. We analyzed the dataset of a laboratory-scale water treatment testbed to illustrate that multi-temporal sequences are applicable for the description of operation patterns. We extended the benchmark simulator of a vinyl acetate production technology to generate easily reproducible results and stimulate the development of alarm management algorithms. The results of detailed sensitivity analyses confirm the benefits of the application of temporal alarm suppression rules which are reflecting the dynamical behaviour of the process.

The files are the supplementary materials of our paper will be published in IEEE Access, 2018 For the extended simulator of the vinyl acetate production technology and the source codes of the Bayes’ theorem-based evaluation of sequences see: HTTPS://GITHUB.COM/ABONYILAB/VACSIMULATOR

The MATLAB implementation of the sequence mining algorithm is available at: HTTPS://GITHUB.COM/ABONYILAB/MULTI-TEMPORAL-SEQUENCE-MINING