Computer system and method for monitoring the technical state of industrial process systems

Patent metadata

Field	Value
Publication number	US-10969774-B2
Application number	US-201916576832-A
Country	US
Kind code	B2
Filing date	Sep 20, 2019
Priority date	Mar 24, 2017
Publication date	Apr 6, 2021
Grant date	Apr 6, 2021

Abstract

An anomaly detection module is configured to apply a plurality of machine learning models to received technical status data to detect one or more indicators of an abnormal technical status prevailing in the industrial process system. The plurality of machine learning models are trained on historic raw or pre-processed sensor data and the anomaly detection module configured to generate the anomaly alert based on the one or more indicators. The received technical status data is assigned to signal groups and the generated anomaly alert is a vector with each vector element representing a group anomaly indicator for the respective signal group. Each vector element is determined by applying a respective group specific machine learning model.

First claim

What is claimed is:

1. A computer system for monitoring an industrial process system under control of an advanced process controller, the industrial process system having an operation part for processing flow materials, the advanced process controller being responsive to one or more sensor signals, the computer system comprising:
one or more processors; and
a non-transitory computer-readable medium having processor-executable instructions stored thereon, wherein the processor-executable instructions, when executed, facilitate:
receiving technical status data describing a current technical state of the industrial process system associated with a respective processing component or a processed flow material, wherein the technical status data corresponds to or is derived from the one or more sensor signals;
assigning the received the technical status data to one or more signal groups, wherein the one or more signal groups are defined by any one of the following grouping criteria:
grouping based on physical location of the respective signals within the industrial process system,
grouping of correlated signals identified by correlation analysis, or
grouping of signals with dependencies as a result of domain knowledge;

applying, based on the assigned one or more signal groups, a plurality of machine learning models to the received technical status data to analyze the technical status data for detecting one or more indicators of an abnormal technical status associated with the industrial process system, wherein the plurality of machine learning models are trained on historic raw or pre-processed sensor data;
generating an anomaly alert based on the one or more indicator, wherein the generated anomaly alert is a vector with each vector element of the vector representing a group anomaly indicator for the respective assigned signal group, wherein each of the vector elements is determined by applying a respective group specific machine learning model from the plurality of machine learning models; and
outputting the anomaly alert, wherein the anomaly alert is configured to enable deactivating of the advanced process controller in case of an anomaly detection for the industrial process system.

2. The computer system of claim 1 , wherein the processor-executable instructions, when executed, further facilitate:
determining one or more root causes within the received technical status data, wherein the one or more root causes indicates an origin for the anomaly alert in the industrial process system, and wherein determining the one or more root causes is based on any one of the following:
density based screening where technical status data samples showing anomalies are compared with samples showing no anomalies to identify the most deviating dimensions as root cause signals; and
scoring the anomaly detection applied to groups of signals and/or single signals to generate a list of most likely root cause signals.

3. The system of claim 1 , wherein the applying the plurality of machine learning models to detect the one or more indicators of the abnormal technical status is based on any one of the following: k nearest neighbor, local outlier detection, fuzzy logic based outlier detection, multi-variate Gaussian distribution, one-class support vector machine, replicator neural networks, self-organizing maps, deviations from association rules, and deviations from frequent item-sets.

4. The system of claim 1 , wherein the processor-executable instructions, when executed, further facilitate:
storing an upper threshold value and a lower threshold value per received technical status data for one or more of the received technical status data;
comparing the received technical status data with the threshold values; and
generating an anomaly indicator for a particular received technical status data based on the technical status data falling outside an interval defined by the upper and lower thresholds.

5. The system of claim 4 , wherein the threshold values are predefined or learned by the respective group specific machine learning model from the historic sensor data.

6. A computer-implemented method for monitoring an industrial process system being under control of an advanced process controller, the advanced process controller being responsive to one or more sensor signals, comprising:
receiving technical status data describing a current technical state of the industrial process system associated with a respective processing component or a processed flow material, wherein the technical status data corresponds to or is derived from the one or more sensor signals;
assigning the received technical status data to one or more signal groups, wherein the one or more signal groups are defined by any one of the following grouping criteria:
grouping based on physical location of the respective signals within the industrial process system,
grouping of correlated signals identified by correlation analysis, or
grouping of signals with dependencies as a result of domain knowledge;

applying, based on the assigned one or more signal groups, a plurality of Machine Learning Models to the received technical status data to analyze the technical status data for detecting one or more indicators of an abnormal technical status associated with the industrial process system, wherein the plurality of Machine Learning Models are trained on historic raw or pre-processed sensor data;
generating an anomaly alert based on the one or more indicators, wherein the anomaly alert is configured to enable deactivating of the advanced process controller, wherein the generated anomaly alert is a vector with each vector element of the vector representing a group anomaly indicator for the respective assigned signal group, wherein each of the vector elements is determined by applying a respective group specific Machine Learning Model from the plurality of machine learning models; and
outputting the anomaly alert to an operator of the industrial process system or to the industrial process system.

7. The method of claim 6 , further comprising:
determining one or more root causes within the received technical status data, wherein the one or more root causes indicates an origin for the anomaly alert in the industrial process system, and wherein determining the one or more root causes is based on any one of the following:
density based screening where sensor data samples showing anomalies are compared with samples showing no anomalies to identify the most deviating dimensions as root cause signals; and
scoring anomaly detection applied to groups of signals and/or single signals to generate a list of most likely root causes.

8. The method of claim 6 , wherein the applying the plurality of machine learning models to detect the one or more indicators of the abnormal technical status is based on any one of the following: k-nearest neighbor, local outlier detection, fuzzy logic based outlier detection, multi-variate Gaussian distribution, one-class support vector machine, replicator neural networks, self-organizing maps, deviations from association rules, and deviations from frequent item-sets.

9. The method of claim 6 , further comprising:
storing an upper threshold value and a lower threshold value per received technical status data for one or more of the received technical status data;
comparing the received technical status data with the threshold values; and
generating an anomaly indicator for a particular received technical status data based on the sensor data falling outside an interval defined by the respective upper and lower thresholds.

10. The method of claim 9 , wherein the threshold values are predefined or learned by the respective group specific M

Assignees

• Abb Schweiz Ag

Inventors

• Hollender Martin
• Kloepper Benjamin
• Lundh Michael
• Chioua Moncef

Classifications

• G05B13/04

  involving the use of models or simulators · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G05B23/0229 Primary

  knowledge based, e.g. expert systems; genetic algorithms · CPC title

• G05B23/024

  Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks · CPC title