Classifications

• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B19/00—Program-control systems
  • G05B19/02—Program-control systems electric
  • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
  • G05B19/41875—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by quality surveillance of production
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B23/00—Testing or monitoring of control systems or parts thereof
  • G05B23/02—Electric testing or monitoring
  • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
  • G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
  • G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
  • 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
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B23/00—Testing or monitoring of control systems or parts thereof
  • G05B23/02—Electric testing or monitoring
  • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
  • G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
  • G05B23/0221—Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B23/00—Testing or monitoring of control systems or parts thereof
  • G05B23/02—Electric testing or monitoring
  • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
  • G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
  • G05B23/0275—Fault isolation and identification, e.g. classify fault; estimate cause or root of failure
  • G05B23/0281—Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B2219/00—Program-control systems
  • G05B2219/30—Nc systems
  • G05B2219/37—Measurements
  • G05B2219/37591—Plant characteristics
• • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
  • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
  • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
  • Y02P80/00—Climate change mitigation technologies for sector-wide applications
  • Y02P80/30—Reducing waste in manufacturing processes; Calculations of released waste quantities
• • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
  • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
  • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
  • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
  • Y02P90/80—Management or planning

Definitions

• the present teachings relate generally to computer-based monitoring of an industrial plant. Background art
• Industrial plants such as process plants comprise equipment that is operated to produce one or more industrial products.
• the equipment may, for example, be machinery and/or heat exchang ers that require maintenance.
• a requirement for maintenance can depend on several factors that include operation time and/or load on the equipment, environmental conditions that the equipment has been exposed to, and so forth.
• Undue or unplanned shutdown of the equipment is generally not desired as it often results in a stoppage of production, which can reduce the efficiency of the plant and can cause wastage. Since the time-period between two mainte nances can vary, it may be difficult to plan the shutdown of the equipment around the time when the maintenance is actually necessary. Accordingly, routine maintenance may either be done earlier than it actually is necessary, or the equipment operation may overrun the maintenance period.
• the latter can affect the life of the equipment, and/or cause operation with poor efficien cy. As it will be appreciated, the latter can also increase the risk of an unplanned shutdown which can cause wastage of materials that could not be processed by the equipment due to the unplanned shutdown. Although the former approach can be used to reduce the risk of an un planned shutdown, the approach may not always be desirable either, as it may result in more frequent maintenances, which can increase costs.
• Plants also comprise a plurality of sensors for measuring or detecting one or more parameters related to the equipment. Some sensors may also require maintenance themselves, for exam ple, preventive maintenance and/or calibration to ensure their reliability in measurement and/or detection of the parameters that they are supposed to measure or detect.
• an output change of a sensor can also indicate the health of the equipment that the sensor is measuring. It is, however, also possible that the output change has been caused by an ill functioning of the sensor itself, and not by reduced health of the equipment. Plants can often comprise several hundreds or thousands of sensors. Large industrial plants can comprise several tens of thousands of sensors, or even more. It can thus be challenging to obtain an indi cation of the equipment health by monitoring each sensor. Moreover, it can be challenging to determine to state of the equipment even if a sensor output drifts. As a result, false positive events may be triggered. Frequent false event signals or alarms may reduce the usability of such a system.
• a method for monitoring a plant comprising a plurality of sensors, and one or more functionally connected processing units, the method comprising: providing, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residue signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitoring, via any of the one or more processing units, a level signal; wherein the level signal is indicative of a collective time-based variation of the time-series residual data, monitoring, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generating, via any of the one or more processing units, an anomaly event signal when at a given time a value of the level signal and/or a value of the
• the last step can also alternatively be expressed as, generating, via any of the one or more processing units, a level event signal and/or an association event signal; wherein the level event signal is generated when at a given time a value of the level signal changes from an expected value of the level signal at or around that time, and the association event signal is generated when at a given time a value of the associa tion signal changes from an expected value of the association signal at or around that time.
• An occurrence of the level event signal and/or an association event signal may be considered by the processing unit as the anomaly event.
• the level event signal and/or an association event, or more generally the anomaly event signal is indicative of an anomaly in at least one of the equipment in the plant. Any one or both of the values may also be time- dependent values.
• time-dependent in this disclosure refers to such values or parameters that can vary with time. Such values may not have a direct depend ence on time, rather that due to time-varying nature of the signals and process parameters re lated to the plant which are time-series data, such values can be represented or computed as a series of discrete or continuous values along a time scale, or time-series values. Accordingly, it will also be appreciated that is not a must that such values always or regularly have to change with the progression of time.
• the last step may even be: generating, via any of the one or more processing units, an anomaly event signal when at a given time a magnitude of the residue signal exceeds beyond a residue threshold, and a value of the level signal and/or the association signal changes from an expected value of the respec tive signal at or around that time.
• the residual data may be provided at the one or more functionally connected processing units as input data.
• the first step may be spe cifically: receiving, at any of the one or more processing units, time-series residual data of a sensor object; the sensor object being a group of at least some of the sensors from the plurality of sen sors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output.
• the residual data may be received directly at one or more inputs of the one or more processing units.
• the residual data may be received at a computer memory functionally con nected to any of the one or more processing units.
• the residual data may even be generated by the one or more functionally con nected processing units.
• the first step may be specifically: generating, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output;
• the method also comprises: monitoring, via any of the one or more processing units, the time-series residual data of a sensor object.
• Monitoring of the residual data may be done by monitoring each residual signal, i.e., for each sensor or any one or more of the residual signals. Accordingly, for any sensor of which the re sidual signal magnitude exceeds a predetermined value or threshold at any given time, a resid ual event may be recorded for that time in a memory location functionally connected to any of the one or more processing units. No action, e.g., raising an alarm or recording an anomaly, is performed if the level signal and the association signal are as expected. Whereas, if the level event signal and/or the association event signal has occurred at or around a given time that the residual event(s) from one or more sensors occurred.
• the residual event signal(s) from one or more sensors may be analyzed to find the source or root cause of the anomaly.
• the residual event signal may have occurred either at the same time, or earlier, or after the occurrence of the level event signal and/or the association event signal.
• the monitoring of the sensor signal(s) may even be based on the measured signal magnitude of any sensor exceeding beyond a given threshold from the expected value of that sensor. Both the cases are hence considered equivalent and may be used interchangeably in this disclosure, for example with reference to thresholds.
• “Sensor object” may refer to a specific group of at least some of the sensors from the plurality of sensors of the plant.
• the sensor object is a group of sensors signals of which residual signals are collectively monitored in the form of the level signal and/or the association signal.
• the number of sensors that are to be monitored are large.
• the number of sensors in a chemical or biological plant can be well above 1000, often there can be several tens of thousands of sen sors, or in some cases hundreds of thousands of sensors or even more.
• chemical and/or bio logical plants with complex production and value chains such a plurality of industrial plants or a Verbund arrangements, the number of sensors can be huge.
• the proposed sensor object allows applying the multivariate techniques disclosed herein more effectively.
• the applicant has realized that if the multivariate techniques are realized to a group that includes all sensors of the plant or a sub-optimal group of sensors, such a group can lack sensitivity with respect to the anomalies to be detected. Smaller but important deviations in some of the sensors can be overcontrolled or dominated by larger but less important deviations in other sensors in the group.
• the present teachings also allow defining suitable groups of sen sors in the form of one or more sensor objects, each of which can allow maintaining sensitivity for anomalies without being overcome by other sensor signals.
• multivariate monitoring methods can have an advantage of reducing false-positive rate in generating alerts that point to an anomaly within the sensor ob ject.
• the sensor object as herein disclosed can allow maintaining low false positives while achieving high sensitivity for anomaly detection.
• Clustering of at some of the sensors from the plurality of sensors into one or more sensor ob jects can be done in many ways as discussed above.
• sub-optimal clustering or group ing of sensors in a sensor object can affect sensitivity of detection.
• manual clustering to generate a sensor object can take a long time without a guarantee of success.
• grouping the sensors manually via manual input can lead to huge man ual workload even if it is to be performed in large scale. Since each plant may be unique in it self, it can be hard to perform a proper grouping as there can be a number of unknowns in volved.
• the information on plant topology is not provided in a processable data format, e.g. mathematical model of a plant that can be processed or evaluated - at least not to the required degree of detail.
• the sensor object is provided by at least partially automatically selecting the at least some of the sensors from the plurality of sensors. Said selection is made based on the suitability of the at least some of the sensors to be grouped within the sensor ob- ject.
• similarity detection using a data-centric algorithm is used for build ing the sensor object. Preferably, the selection is done fully automatically.
• the selecting of the at least some of the sensors, or grouping of the sensors is per formed by one or more data-centric methods or algorithms.
• the data-centric algo rithm is configured to use at least one similarity measure for grouping sensors from the plurality sensors into sensor objects.
• the data-centric algorithm uses at least one similarity measure to group the at least some of the sensors for providing the sensor object.
• da ta-centric algorithm it is meant an algorithm that is configured to leverage sensor data, such as historical time-series data of the plurality of sensors, for at least partially automatically grouping or selecting the at least some of the sensors.
• the data-based algorithm may be one or more clustering algorithms.
• the clustering algorithm may be an unsupervised learning algorithm such as a neural network model trained via unsupervised learning using the historical data of the plurality of sensors, or any other suitable algorithm for clustering and dimension re duction.
• the unsupervised learning algorithm may, for example, be self-organizing maps (“SOM”). The applicant has found SOM to be particularly beneficial for automatically grouping appropriate sensors into sensor objects.
• any one or more of the processing units is configured to arrange or order, at a computer memory, the sensors accord ing to similar patterns, or one or more similarity measures, in their time-dependent signals.
• historical time-series data from the corresponding sensors can be leveraged.
• the sensors may be arranged on a matrix or a computer-readable 2D map space.
• this matrix or map is subdivided or fragmented for generating one or more sensor objects.
• the map space may be fragmented or cut using a symmetrical or asymmetrical grid. Additionally, or alternatively, the map space is cut using a distance value around a cluster of sensors on the map.
• a cluster may be detected automatically via a similarity measure that involves selecting the sensors that lie within a certain distance from each other. Then all sensors that lie within a distance value may be grouped in a sensor object.
• the dis tance value measure may even comprise multiple distance values, for example for capturing sensors that form an asymmetrical cluster.
• sub-clusters within a cluster may be detected based on one or more distances for better capturing asymmetric clusters.
• the entire plurality of sensors is divided as explained into sensor objects.
• the sensor object may have two or more sensors. In a preferred aspect, the sensor object may have 20 or few sensors, but not below 2 sensors.
• the population of the sensors within the sensor objects may differ, e.g., some sensor objects may have more than 20 sensors, for example around 100 sensors...
• the sensor object comprises signals from 20 or around 20 sensors.
• the sensor object comprises signals from 10, around 10, or several tens of sensors.
• the term “several tens of sensors” here is meant to include any integer number of sensors equal to or less than 100, for example, 4, or 12, or 25, or 30 sensors.
• Proposed automatic clustering allows capturing the sensors that are suitable to be included within a sensor object, e.g., by allowing automatic detection of similarity between the sensors.
• the anomaly detection can be performed without an expert user and with little or no specifics of the sensors or the plant topology. For complex and large plants such as chemical and biological plant this can be a significant advantage.
• the SOM is trained using unsupervised learning using sensor data from the plurality of sensors to produce a two-dimensional, discretized representation of the input vectors, which in this case would be the sensor data that needs to be clustered into sensor ob jects, i.e., data from the plurality of sensors.
• those of the in put vectors, or sensor signals, that are similar in high dimensional space are mapped to nearby nodes in a two-dimensional (“2D”) space.
• the similarity may be measured in terms of distance between the sensor nodes mapped in the 2D space.
• the 2D space may be defined or specified for example, beforehand by representing its geometry as a A* n grid.
• the sensor nodes may be mapped into the 2D space, initially at random, and then their position is iteratively ad justed. In this way, neighboring points in the initial geometry of the input vectors can be mapped to nearby points in the 2D space.
• one or more SOMs can be used to cluster times-series into groups with matching shapes.
• groups represent sen sor objects.
• processing units need not be located at the same site or at the same physical location.
• at least some of the processing units may be implement ed as, or at, a cloud service. Since each of the processing functions of monitoring the level sig nal and the association signal provide at least one technical advantage, such functions are pa tentable also in their own right.
• a method for monitoring a plant comprising a plurality of sensors, and one or more functionally connected processing units, the method comprising: providing, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitoring, via any of the one or more processing units, a level signal; wherein the level signal is indicative of a collective time-based variation of the time-series residual data, generating, via any of the one or more processing units, a level event signal; wherein the level event signal is generated when at a given time a value of the level signal changes from an expected value of the level signal at or around that time, wherein the level event signal is indica tive of an anomaly in at least one of the
• a method for monitoring a plant comprising a plurality of sensors, and one or more functionally connected processing units, the method comprising: providing, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitoring, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generating, via any of the one or more processing units, an association event signal; wherein the association event signal is generated when at a given time a value of the level sig nal changes from an expected value of the association signal at or around that time, wherein the association event signal is indicative of an anomaly in at least one of the
• the sensor object is provided by at least partially automatically grouping the at least some of the sensors.
• the present teachings defer the anomaly event signal generation based on the monitoring of the level signal and/or the association signal.
• the anomaly event is generated when either one or both of the level signal and the association signal change from their respective expected value. Accordingly, in other words, the generation of the anomaly event signal is prevented when any of the residual signal exceeds beyond their residue threshold while both the level signal and the association signal are within their respective expected values, anomaly event is thus based on the result of level signal and/or association signal monitoring.
• the respective expected values of the level signal and the association signal are each preferably provided as a respective range of values within which these respec tive signals may validly lie.
• Each of the level signal and the association signal may thus be pro vided at least one limit value at any given time. If the signal value is within the respective limit at that time, the value is considered as expected and anomaly event is not generated.
• any of the respective expected value may be provided as a corresponding expected val ue limit, specifying for a given time a plurality of expected values that the corresponding signal may validly have without the anomaly event being generated.
• the limit values may also be time- dependent values.
• the plurality of expected values may either be dis crete values, or they may correspond to any value the respective signal can take within the range specified by the corresponding limit value.
• the applicant has found that the proposed event generation based on the monitoring of the lev el and association signals can result in significant reduction of false positive events whilst pre serving focus on detecting a real anomaly.
• an alarm may be generated to inform an operator or user about an anomaly in an equipment that the sensor object is related to.
• any of the one or more processing units henceforth simply termed as the processing unit, may backtrack the sensor data to determine the source of the anomaly.
• the method also comprises: determining, in response to the anomaly event signal, health of an equipment related to the sensor object.
• the health of the equipment may be determined, for example, by performing a root cause anal ysis via the processing unit.
• the change or deviation from the expected value is provided with a limit value. If the signal value lies within its associated limit value, no event signal is gen erated. Accordingly, when any of the monitored signals, level or association, of the sensor ob ject deviates beyond a given limit, or control range, this deviation can be considered as an anomaly and an alert may be generated in the form of the event signal, (e.g., the level event signal, the association event signal, or both). As proposed, the state of the sensor object is monitored or observed using the level signal and/or the association signal, both being time- dependent or time-based signals.
• the event signal e.g., the level event signal, the association event signal, or both.
• the given limits can be defined as per the application of interest, for example, the acceptable tolerance of deviation, required sen sitivity of events, criticality or importance of the sensor object, and similar.
• the limit values are derived using statistical limits within which the respective signals, level and association signals, are expected to lie under normal conditions.
• the limit value thus repre sents a value range or probability space for the movement of, or around, the expected value within which the monitored signal value can be validly located.
• the proba bility space which lies within the upper limit value and the lower limit value may thus be defined by a plurality of expected values.
• the probability space can be determined using historical data, e.g., from past monitoring.
• the limit values may thus be understood as defining a value range within which range the observed value may be allowably located with a given probability value.
• a limit value can, as per application requirements, and/or availability of the historical sensor data, either be zero or a non-zero value.
• the limit value may also be a time-dependent value. Accordingly, it is not essential to attach a specific number to the limit values in this disclosure.
• the two indicators discussed above i.e., the level signal and the association signal
• the indicators are applied to the residual signals of the sen- sor object.
• Residual signal which is generated for each sensor of the sensor object, is a differ ence between the measured sensor output signal (or observed sensor signal) at a given time and an expected sensor signal at that time.
• the indicators are then generated from the residual data which comprises the residual signals from the sensor object sensors.
• the measured sensor output signal in most cases comprises a plurality of infor mation, most of which can be irrelevant for detecting an anomaly, or equipment health.
• the measured or observed sensor output signal can depend upon controller settings, production mode, operating conditions of the plant and or equipment, etc.
• the information relat ed to the equipment health can thus drown within superfluous information caused by the various other parameters that the sensor output is dependent upon.
• the applicant has realized that ra ther than using sensor output signals directly, by applying the proposed indicators on the resid ual data instead, the superfluous information can at least partially be removed, from the time dependent sensor outputs, such that the health-related information of the equipment can be come more detectable for further signal processing.
• the first indicator provides information or a statistical value related to the col lective movement of the time dependent residual signals in the time-series residual data of the sensor object.
• the level signal value can be used to detect level changes or short-term trends of the time-series residual data within the sensor object.
• the second indicator, the association signal provides information or a statistical value related to the variation and/or association structure of the time-series residual data.
• the association signal is indicative of changes in vola tility and/or correlation structure among the time-series residual signals within the sensor object.
• the event signal for the respective signal, the level signal and/or the association signal can be generated when the magnitude of the signal value goes beyond an expected value or a given limit of the expected values of that signal at that time.
• the indicators are thus compared at any given time with respect to, what can be called an expected state thereof for that time.
• the ex pected state or time dependent expected signal values can be provided by a model of the sen sor object.
• the sensor object model can be a predictive model, such as at least partially a data- driven model, e.g., comprising a sensor object neural network, that has been trained using his torical residual data.
• the results of the comparison i.e., a deviation of the level signal from its expected state, and a deviation of the association signal from its expected state, are monitored via the processing unit over time.
• the anomaly event signal is generated when the magnitude of the level signal value and/or the association signal value diverges, beyond a limit, from the ex pected value of that signal at that time.
• the deviations are monitored via the processing unit over time.
• Monitoring of the signals or even the deviations can either be done continuously or it can be done between discrete time periods, of either equal or unequal lengths.
• the sensor object can be provided at the processing unit, for example, via a memory functional ly connected to the processing unit.
• generation of the residual data is done via the same processing unit.
• the generation of the residual data can be done by another processor and then provided at the processing unit.
• Industrial plants or simply plants, comprise infrastructure that is used for an industrial purpose.
• the industrial purpose may be manufacturing or processing of one or more products, i.e., a manufacturing process or a processing performed by the plant.
• the product can, for example, be any physical product, such as a: chemical, biological, pharmaceutical, food, beverage, tex tile, metal, plastic, semiconductor, or the product can even be a service product such as: elec tricity, heating, air-conditioning, waste treatment such as recycling, chemical treatment such as breakdown or dissolution, or even incineration, etc.
• the plant can be any or more of the: chemical plant, pharmaceutical plant, fossil fuel processing facility such as oil and/or natural gas well, refinery, petrochemical plant, cracking plant, and such.
• the plant can even be any of the: distillery, incinerator, or power plant.
• the plant can even be a combination of any of the above, for example, the plant may be a chemical plant that includes a cracking facility such as a steam cracker, and/or a power plant.
• a sub facility within a large plant may even be considered a plant.
• the infrastructure can comprise equipment or process units such as any one or more of: heat exchanger, column such as frac tionating column, furnace, reaction chamber, cracking unit, storage tank, precipitator, pipeline, stack, filter, valve, actuator, transformer, circuit breaker, machinery e.g., heavy duty rotating equipment such as turbine, generator, pulverizer, compressor, fan, pump, motor, etc.
• the plant or industrial plant may even be part of a plurality of industrial plants.
• the term “plurali ty of industrial plants” as used herein is a broad term and is to be given its ordinary and cus tomary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning.
• the term specifically may refer, without limitation, to a compound of at least two industrial plants having at least one common industrial purpose.
• the plu rality of industrial plants may comprise at least two, at least five, at least ten or even more in dustrial plants being physically and/or chemically coupled.
• the plurality of industrial plants may be coupled such that the industrial plants forming the plurality of industrial plants may share one or more of their value chains, educts and/or products.
• the plurality of industrial plants may also be referred to as a compound, a compound site, a Verbund or a Verbund site.
• the value chain production of the plurality of industrial plants via various intermediate products to an end product may be decentralized in various locations, such as in various industrial plants, or inte grated in the Verbund site or a chemical park.
• Such Verbund sites or chemical parks may be or may comprise one or more industrial plants, where products manufactured in the at least one industrial plant can serve as a feedstock for another industrial plant.
• the plant usually also comprises instrumentation that can include several different types of sensors. Sensors are used for measuring various process parameters and for measuring parameters related to the equipment. For example, sensors may be used for measuring process parameters such as flowrate within a pipeline, level inside a tank, temperature of a furnace, chemical composition of a gas, etc., and some sensors can be used for measuring vibration of a turbine, speed of a fan, opening of a valve, corrosion of a pipeline, voltage across a transformer, etc. The difference between these sensors can not only be based on the parameter that they sense, but it may even be the sensing principle that the respective sensor uses.
• sensors based on the parameter that they sense are: temperature sensors, pressure sensors, radiation sensors such as light sensors, flow sen sors, vibration sensors, displacement sensors and chemical sensors, such as those for detect ing a specific matter such as a gas.
• sensors that differ in terms of the sensing prin ciple that they employ are for example: piezoelectric sensors, piezoresistive sensors, thermo couples, impedance sensors such as capacitive sensors and resistive sensors, and so forth.
• the plant is thus often equipped with sensors that measure the value of a certain quantity in the plant (e.g., temperature in a column, pressure in a pipe, mass flow, etc.) either continuously or at regular intervals.
• a certain quantity in the plant e.g., temperature in a column, pressure in a pipe, mass flow, etc.
• PI MS plant information system
• the additional information may be one or more of the metadata such as: sensor name, timestamp of measurement, value of measurement, unit, quality of measurement.
• a control sys tem that controls at least some of the operations of the plant.
• Some plants may even comprise multiple control systems that may be configured to operate in a hierarchy, or in parallel. Exact architecture of the one or more control systems, Industrial Control Systems (“ICS”), is not es sential for the scope of generality of the present teachings. It is also usual for plants to have a data acquisition system (“DAS”) that receives data from a plurality of sensors within the plant.
• the sensor data is stored in a long-term computer memory or a database. DAS may be the same system as PI MS or they may be different systems.
• the sensor data usually includes metadata indicating the history or time information of the data collected from the sensors.
• the historical sensor data is hence usually stored in, or is recoverable from, the database as a time- series data. Historical association and/or level signal data may also be stored in the same data base or another database functionally connected to the processing unit.
• the metadata can also include unit and/or label of the sensor time-series data.
• a sensor is often also termed a tag or a sensor tag in industrial plants.
• plant control and/or monitoring systems are: Programmable Logic Controller (“PLC”), Distributed Control Sys tem (“DCS”), and Supervisory Control and Data Acquisition (“SCADA”).
• PLC Programmable Logic Controller
• DCS Distributed Control Sys tem
• SCADA Supervisory Control and Data Acquisition
• the function of any two or more of the systems mentioned above may be performed by a single control and/or monitoring system.
• the applicant has realized that by the proposed method or the system for monitoring thereof, it can be made possible to detect an onset of an anomaly or unusual behavior in the plant at an early stage before the problem becomes observable by an operator or a conventional monitor ing system. It can thus be prevented that at least some of the plant equipment unexpectedly shuts down due to an anomaly that manifested itself over time. Such prevention can be made by detecting the anomaly using the present teachings and planning a maintenance such that disruption of the industrial process can be at least reduced. It can also be provided to guide the operator’s attention to a specific plant area or equipment where the problem may later manifest itself such that a remedial action may be planned.
• the pro cessing unit determines a state of health of at least one piece of equipment in the plant.
• the processing unit initiates further analysis involving analyzing historical and/or real-time time-series data of at least one individual sensor or a sub-group of sensors within the plurality of sensors.
• the processing unit prioritizes the further analysis of one or more sensors outputs for which the residue signal exceeded beyond their respective residue threshold, or the at least one sensor which have each an associated residue event signal that occurred around the same time as the occurrence of the anomaly signal.
• the processing unit can determine more specifically which piece of equipment or sensor may require maintenance. More preferably, by analyzing said time-series signals over a given time-period, the processing unit may forecast the maintenance schedule of the at least one piece of equipment. As it will be appreciated, by doing so the pro cessing unit may provide the future maintenance requirements related to the equipment and/or one or more sensors. In addition, the amount of sensor data that the processing unit is required to monitor and analyze in real-time can be reduced by monitoring the proposed level and asso ciation signals.
• the resources used for monitoring can be significantly reduced as com pared to the case when each sensor signal is individually analyzed. As mentioned previously, this can also have additional advantages when it comes to false detections as compared to a univariate approach, i.e., monitoring and analyzing each sensor individually.
• the processing unit analyzes time series residual signal of each sensor within the sensor object to determine one or more main drivers or most dominant contributors to the level signal value.
• the determination may be done, for example, using an effect size calculation and/or value distribution analysis.
• Effect size is a measure of distance of how far a residual signal is from a specific residual value within a given time-period.
• the specific residual value is usually the most probable value of that residual signal within the given time-period.
• the processing unit in response to the association event signal, the processing unit analyz es time series residual signal of each sensor within the sensor object to determine one or more main drivers or most dominant contributors to the association signal value.
• the processing unit analyzes covariance of time series residual signals of each pair combination of sensor residual signals within the sensor object to determine one or more main drivers or most dominant contributors to the association signal value.
• the proposed teachings can leverage the similarity and/or association between the residual data from the sensors within the sensor object to focus computational monitoring effort on the scenarios in which data from one or more sensors deviates in such a way that it has a bearing on the overall level and association of the residual data within the sensor object.
• the grouping of sensor residual data allows monitoring according to the proposed multivariate approach. It will be appreciated that this can free up computational resources while retaining focus on the overall behavior of the relevant sensor data comprised within the sensor object.
• the grouping is at least partially done automatically via the processing unit, for example, using self-organizing maps.
• the automatic grouping may be done by the processing unit based on data characteristics, for example, similarity of the sen sor time-series data of the sensor object and/or based on the interdependency between the output signals of the sensors, and/or type of sensor, and/or similarly in sensor response.
• the grouping is done based on an input from a user. Accordingly, the grouping may be done at least partially based on operator preference or experience.
• the proposed multivariate approach using the level signal and the association has an advantage that interdependencies among the resid uals may be used to detect behavior that may otherwise be undetectable on individual signal level.
• false-detections can be reduced as compared to a univariate monitoring approach, i.e., monitoring data from each sensor individually, whilst retaining sensi tivity of detection of anomalies within the sensor object.
• a method for monitor ing a plant comprising a plurality of sensors, and one or more functionally connected processing units, the method comprising: providing, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residue signal which is a difference between the sensor’s measured output and the sensor’s expected output, wherein the sensor object is provided by at least partially automatically group ing the at least some of the sensors, monitoring, via any of the one or more processing units, a level signal; wherein the level signal is indicative of a collective time-based variation of the time-series residual data, monitoring, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generating, via any of the one or more processing units,
• the automatically grouping is performed using at least one data-centric algorithm.
• the data-centric algorithm may be a clustering algorithm, for example SOM algo rithm.
• a fully automatic grouping is initially performed, and then a part of the grouping is rearranged as per user input. This can be helpful in onboarding a new plant, and manual intervention can be reduced.
• a first group of sensor outputs from a first group of sensors within the plurality of sensors, are configured or tagged as covariate signals. Further, a second group of sensor outputs, different from the first group of sensor outputs, are configured or tagged as monitored signals.
• the sensor object is realized using the residuals from the second group of sensor outputs.
• the first group of outputs, or the covariate signals are the signals that repre sent parameters that can cause a change in the behavior of at least one of the monitored sig nals.
• the covariate signals hence represent influential factors or parameters that preferably can influence the monitored signals.
• the covariate signals may represent parameters such as, am bient temperature, coolant temperature, load, input flow, output flow, controlled measurement or setscrew position.
• One or more or the monitored signals are hence preferably at least partially dependent upon at least one of the covariate signals.
• the processing unit may automatically determine the covariate signals from the monitored signals by checking the interdependencies between the sensor outputs.
• at least some of the covariates and/or monitored sig nals may be defined based on user input. The advantage of consolidated monitoring of sensor data can still be maintained.
• data from the first group of outputs or covariate signals is also input to the sensor object model for generating an expected level signal and an expected association signal and/or the respective limits thereof.
• time dependent values of the level signal and the association signal are compared with the time dependent values of the expected level signal and the expected association signal.
• the applicant has thus realized that the expected values and their limits can be made more accurate by considering the covariates, which can synergisti- cally improve the detection of an anomaly at an early stage while automatically accounting for the factors that can influence the sensor data.
• the expected values can thus also be adapted according to covariate signals. This when combined with the monitoring being done at sensor object level can also result in a solution that required reduced monitoring resources and re Jerusalem positives.
• the sensor s expected output is provided by an expected state model, a Machine Learning (“ML”) model.
• the expected state model is a prediction model, e.g., at least partially a data-driven model, such as an expected state neural network, that is trained using training data that comprises historical time series output data of the sensor.
• the expected state model may be, or it may comprise as a machine learning (“ML”) mod ule, a prediction model that when trained using training data that comprises historical time se ries output data of the sensor results in a trained data driven model.
• ML machine learning
• Data driven model refers to a model that is at least partially derived from data, in this case from the user training data that may comprise historical data related to the sensor.
• a data driven model can allow describing relations that cannot be modelled by physio-chemical laws.
• the use of data driven models can allow to de scribe relations without solving equations from physio-chemical laws, e.g., related to the pro Deads taking place within the respective production process. This can reduce computational power and/or improve speed.
• the data driven model may be a regression model.
• the data driven model may be a mathemat ical model.
• the mathematical model may describe the relation between provided performance properties and determined performance properties as a function.
• the data-driven model preferably data-driven machine learning (“ML”) model or a merely data-driven model
• ML machine learning
• An untrained mathematical model refers to a model that does not reflect reaction kinetics or physio-chemical processes, e.g. the untrained mathematical model is not derived from physical law providing a scientific generalization based upon empirical observation. Hence, the kinetic or physio-chemical properties may not be inherent to the un trained mathematical model. The untrained model does not reflect such properties.
• Feature en gineering and training with the respective training data sets enable parametrization of the un trained mathematical model.
• the result of such training is a merely data-driven model, prefera bly data-driven ML model, which as a result of the training process, preferably solely as a result of the training process, reflects reaction kinetics or physio-chemical processes related to the respective plant and/or one or more of the plant’s equipment or assets.
• the expected state model may even be a hybrid model.
• a hybrid model may refer to a model that comprises first-principles parts, so called white box, as well as data-driven parts as ex plained previously, so called black box.
• the expected state model may comprise a combination of a white-box-model and a black-box-model and/or a grey-box-model.
• the white-box-model may be based on physio-chemical laws, for example expressed as equations.
• the physio- chemical laws may be derived from first principles.
• the physio-chemical laws may comprise one or more of chemical kinetics, conservation laws of mass, momentum and energy, particle popu lation in arbitrary dimension.
• the white-box-model may be selected according to the physio- chemical laws that govern the respective plant, its production process or parts thereof.
• the black-box-model may be based on historical data, such as the historical time series output data of the sensor.
• the black-box-model may be built by using one or more of machine learning, deep learning, neural networks, or other form of artificial intelligence.
• the black-box-model may be any model that yields a good fit between the training data set and test data.
• the grey-box- model is a model that combines a partial theoretical structure with data to complete the model.
• the trained model may comprise a serial or parallel architecture.
• serial architecture out put of the white-box-model may be used as input for the black-box-model or output of the black- box-model may be used as input for the white-box-model.
• parallel architecture a com bined output of the white-box-model and the black-box-model may be determined such as by superposition of the outputs.
• a first sub-model may predict at least one of the performance parameters and/or at least some of the control settings based on a hy brid model with an analytical white-box-model and a data-driven model that serves as a black box corrector trained on the respective historical data.
• This first sub-model may have a serial architecture, wherein the output of the white-box-model is input for the black-box-model, or the first sub-model may have parallel architecture. Predicted output of the white-box model may be compared with a test data set comprising a part of historical data. An error between the com puted white-box output and test data can be learned by the data-driven model and can then applied for arbitrary predictions.
• the second sub-model may have a parallel architecture. Other examples can be possible, too.
• machine learning may refer to a statistical method that ena bles machines to “learn” tasks from data without explicitly programming.
• Machine learning tech niques may comprise “traditional machine learning” — the workflow in which one manually se lects features and then trains the model. Examples of traditional machine learning techniques may include decision trees, support vector machines, and ensemble methods.
• the data driven model may comprises a data driven deep learning model. Deep learning is a subset of machine learning modeled loosely on the neural pathways of the human brain. Deep refers to the multiple layers between the input and output layers. In deep learning, the algorithm automatically learns what features are useful. Examples of deep learning techniques may in clude convolutional neural networks (“CNNs”), recurrent neural networks such as long short term memory (“LSTM”), and deep Q networks.
• CNNs convolutional neural networks
• LSTM long short term memory
• the sensor object model may be, or it may comprise as a sensor object machine learning (“ML”) module, a prediction model that when trained training data that comprises historical time-series data results in a trained sensor object data driven model.
• ML sensor object machine learning
• the expected output of the sensor is thus generated by inputting data from the first group of outputs or at least one of the covariate signals to the expected state model.
• each sensor or tag may be provided with a respective expected state model that has been trained using the historical data from that specific sensor.
• the covariate signals input to the expected state model can either be all the covariate signals of the plant, or they may be a subset of the plant covariate signals.
• the plant can be viewed as a closed system with expected as well as unexpected interdependencies between different equipment. For example, external temperature of an operative furnace may result in that the local ambient temperature is higher than other parts of the plant.
• the covariate input signals to the expected state model and/or the sensor object model are a subset of the all covariate signals of the plant. The latter can save processing pow er that is being used by the expected state model. The expected state model and/or the sensor object model can thus also be made faster.
• the processing unit may determine covariates that are dominant for the respective model.
• the dominant covariates which are preferably a subset of all the covariates, can be determined via the processing unit by analyzing the predictive pow er of each covariate signal on the respective model. Accordingly, if a variation in a covariate does not influence the model output, the covariate is prevented as an input to the model.
• the processing unit may use historical time-series data for analyzing the predictive power of each covariate signal of the plant on each monitored signal. The processing unit can thus determine the expected state model and/or the sensor object model with a subset of selected covariate inputs that observably influences the model output.
• the covariates may be analyzed not only by considering respective sensor data that has oc curred at the same or around the same time, but also by analyzing them with an additional time lag.
• the time lag may be one or more time periods between an occurrence in a covariate and detecting for an effect of that occurrence as a given sensor’s output. This allows capturing the interdependencies that are associated with a delay or a time-constant. As an example, if at time t, fuel input to a furnace was increased, a temperature rise in the liquid heated by the furnace may only have been detected at time t+ t d . Where, t d represents a time lag in the system. The applicant has realized that this can allow the most important covariate signals to be determined even though they might have been falsely ignored, were the effect of such a covariate not ana lyzed with the lag.
• the processing unit selects the type of expected state model by analyz ing which model type provides the lowest error between the expected output and the actual out put.
• the error can be measured by calculating, between the expected output and the actual output, any one or more of: an absolute error value, a mean-square error value, a weighted mean-square error value, or even their combination.
• the expected state model may be trained, for example, with a specific portion of the historical time-series data, and the error may be calculated by applying the trained expected state model on the covariate signal data from another portion of the time series data. The output of the trained model in response to this covariate signal data may be compared with the actual output in the historical data to calculate the error.
• the processing unit may evaluate a plurality of model types each based on a different prediction method and then pick the model which pro vides the lowest error.
• the model with a given accuracy performance score may be selected.
• the score may be a figure of merit (“FOM”), such as the lowest “(mean absolute error) * (processing resources)”.
• FOM may be generated from other types of error or other metrics as well.
• Processing resources may be indicative of processing time, ener gy, or their combination used by the processing unit for performing the prediction functions with the model.
• the weighted mean square error may for example be calculated by assigning different weights to one or more different sections of the training data. This has an advantage that model accura cy can be improved by focusing the model behavior around the one or more sections of the time-series historical data that reflect the sensor behavior more accurately.
• the determination of the model type may usually be done as an initial step.
• the error and/or FOM analyses can be done based on the historical sensor data.
• One or more ex pected state models that are under evaluation may be trained by specifying a time window on the historical time series data.
• the trained expected state models(s) may then be compared by the processing unit for error and/or performance, for example, by using the rest of the historical time series data for the sensor.
• the processing unit may also determine the covariate signals to be used as input to the model. As discussed, these may either be all covariates of the plant, or a subset thereof.
• the subset may either be at least partially specified by a user, or as explained earlier, the processing unit may at least partially select the dominant covariates based on the predictive power of each of the covariates on the model output. The processing unit can then select the best suited expected state model for generating the sensor’s expected output with current time series data from the sensor.
• the training data for training the sensor expected state model contains sensor data related to the normal operating conditions. Training based on undesired operating conditions can be prevented or reduced in this case. By doing so, undesired deviations in the sensor data can be better captured.
• the sensor object model is also preferably also trained with residual data related to the normal operating conditions. This, thus, can have a synergistic effect when used with the proposed sensor object, i.e., monitoring of level and association signals, by condensing the number of parameters to be monitored whilst an improved visibility of undesired variations in the sensor data caused by an anomaly. For the expected state model as well, this can improve the visibility of an abnormal sensor output.
• the plurality of sensors is subdivided into categories such as sensors belonging to a plurality of plant areas.
• the sensors belonging to each plant area, or plant area sensors may be subdivided into sensors belonging to a plurality of process groups.
• the sensors belonging to each process group, or process group sensors may be subdivided into sensors belonging to a plurality of sensor objects.
• each sensor object is realized by grouping the time-series output data from the sen sors belonging that sensor object.
• the sensors or tags belonging thereof are configured either as covariates, or as tags to be monitored.
• the distinction between a covariate tag and a monitored tag is mutually exclusive within the same process group, however a moni tored tag in a process group can also be a covariate in another process group.
• the subdivision into sensor objects is done based on the sensor data that is to be monitored simultaneously.
• Level signal and association signal are thus the indicators that are monitored for each sensor object, as explained earlier.
• the plant is a thermal power plant
• the plurality of sensors in the thermal power plant can be subdivided or tagged according to plant areas such as: water reservoir, generation unit, switch yard, etc.
• the generation unit area may be subdivided into process groups such as: boiler, feedwater loop, turbine, condenser, generator, etc.
• Monitoring object or sensor object may be created either from sensors belonging to the same process group, or it may be created from sensors belong ing to different process groups.
• Such subdivision of the plant can also help in reducing the number of covariates that are relevant for the prediction using the expected state model and/or the sensor object model.
• the covariates that are located close to the ob served sensor or the sensor object can be considered.
• the proposed teachings provide two distinct states for the sensor, i.e., an actual state, which represents the observed or meas ured output value of that individual sensor at any given time t, and an expected state which rep resents an expected value for that sensor at the time /.
• the expected state is preferably defined from the normal plant operation mode or operation.
• the actual state of the sensor is compared with the expected state of the sensor to generate the residual signal for that sensor.
• the residu al signal thus represents the deviation of the actual or observed state from the expected state of the sensor.
• the level signal value changes beyond an expected level signal value or a given level signal limit
• the association signal value changes beyond an expected association signal value or a given asso ciation signal limit
• the association event signal is generated.
• the limit value can either be an absolute value or it may be a value relative to the respective signal val ue.
• the thresholds and limits can be defined as per application. According to another as pect, one or both thresholds for a residual signal is determined by the processing unit using a control chart.
• one or both limits for either one or both of the level signal and the association signal are determined by the processing unit using a respective control chart.
• the control chart is generated using the respective resid ual or score data.
• the limits are defined as a quantile value, for example of the expected values of the respective score.
• the upper control limit may be 99.5% quantile or thereabouts.
• the lower control limit may be 0.5% quantile or thereabouts.
• the expected state model and/or the sensor object model is retrained between one or more predetermined time intervals.
• expected state determination can be correctively recalibrated by taking into account a change in response of the sensor(s) that may be caused by natural factors such as ageing and/or drift in calibration, etc.
• the model retraining may also be triggered automatically in response to the model performance going be low a minimum performance threshold of the model.
• the sensor ob ject model retraining may be automatically triggered in response to a change in the process parameters, such as a user input to change the plant output, for example increase or decrease in production. The retraining can thus capture the change in the behavior of the equipment and/or the plant.
• the level signal value is generated using a distance estimator, for ex ample, the T2-Hotelling statistic on the residual data of the sensor object.
• the T2-Hotelling is a generalization of the t-statistic and indicates deviations from the multivariate mean of a group of variables. In general, the higher the value of the T2-Hotelling statistic, the more distant is an observation from the mean.
• the association signal value is generated using a measure of multivariate dependencies, for example, the Generalized Variance (“GV”) statistics on the residual data of the sensor object.
• GV Generalized Variance
• the GV may be calculated as the determinant of the variance-covariance matrix of a sample of observations and is a multivariate generalization of the variance. It can therefore be used to measure at a given time the dispersion of time-series residual data in a sensor object.
• Each of the signal values are calculated for time-series residual data from a respective time window, each being of a specific length.
• the time window can be selected based, for example, on the quality of the training data.
• historical level signal and/or association signal are recorded as a time series data on a database functionally connected to the processing unit.
• any of the time series data (e.g., the sensor time series data and/or the residual data and/or the level signal data and/or the association signal data) also comprise an notation data.
• Annotation data can be provided via a user input, but in some cases it may even be at least partially automatically provided by the processing unit.
• the annotation data may be provided with types and/or levels.
• the annotation type may classify features of the data such as what a certain section of the time series data relates to, for example, maintenance activity, breakdown, data issue, etc.
• the level type may specify for which level in plant the annotation relates to.
• a plant level may be used to refer to an annotation that relates to all sensor objects and their sensors
• an area level may refer to an annotation relating to all sensor objects and their sensors that belong to a specific plant area
• similarly process group level sen sor object level and even sensor level annotations may be specified.
• the processing unit automatically selects, using one or more annotations, a suitable portion of historical data for training the expected state model and/or the sensor object model.
• the processing unit may avoid historical time windows that include a certain type of annota tion within the time series. By doing so, it can be avoided that the models and trained on incon sistent data. Model accuracy can thus be improved by selecting training data that provides proper information on normal plant or equipment operation.
• the pro cessing unit automatically places annotations to define a historical time window according to desired limit(s) and/or threshold(s) for the score(s) and/or residual(s) respectively. This can thus be used to influence how tightly the observed signal(s) need to track the expected signal(s).
• the user specified annotations may be received with a user interface (“Ul”) and, for example, stored in an analysis database.
• the annotations may even include timestamp(s) for specifying the start and end of the annotation.
• the annotations may even be used to retrieve the desired sections of the time series data.
• creeping processes such as fouling of a pipe, may cause slow trends or drift in the sensor data.
• Such trends may cause slow rise or droop in the sensor data values over time.
• the trends may have a small slope or rate of change, which may result in that a given detecta ble change in the sensor output caused by such a creeping process takes for example, a week, several weeks or even months to appear.
• a retraining of the expected state model and/or the sensor object model can cause such slow trends to become undetectable by monitoring the level and/or association signals, as well as residuals.
• the processing unit may perform a trend detection. The applicant has found the calculation of the strength, smoothness and currentness of the historical data of the sensor can be particularly useful in detecting the drift in the time series data.
• the method can also comprise: detecting, via any of the one or more processing units, a drift in the output signal of a sen sor; wherein the sensor is among the at least some of the sensors, and wherein the drift is computed from the historical time series data of the sensor, the historical data of the sensor being from a time period that is at least 1 week long, and wherein the drift is detected by com puting the strength, smoothness and currentness of the historical data of the sensor.
• strength in this context refers to strength of the sig nal trend.
• the strength may thus be represented via a measure for the slope of the trend.
• the strength describes the intensity and recognizability of the drift that can be measured, for exam- pie, using a Mann-Kendall test on the historical data of the sensor. Such a test results in a value score that can indicate whether a trend can be detected and whether it is a positive or a nega tive trend.
• a measure of the strength can thus indicate whether the trend or drift is weak or strong.
• the smoothness describes whether the drift is rather smooth or if it is caused by a more abrupt level shift in the historical data.
• the smoothness thus represents standardized measurement of the degree of absence of abrupt features in the data, such as spikes, level-shifts etc. Presence of such features may add to the uncertainty in the trend, for example, the strength of the trend may be artificially inflated or deflated.
• the present teachings thus propose using the smooth ness value in context of the strength and currentness to detect or compute the drift in the output signal the sensor, which will be closer to the actual drift in the sensor output.
• the above criteria can synergistically allow detecting actual drifts that can have bearing into anomalies or potential anomalies. This can thus allow preserving such actual drifts whilst ignoring such variations in the sensor output that do not represent or are not likely to represent anomalies. Thus, despite retraining of the expected state model, a more reliable visibility can be maintained on the more slow-moving effects that have not manifested yet as anomalies.
• the time period for computing drift is a duration that is 1 month or around 1 month long. Additionally or alternatively, according to another aspect, the duration is 3 months or around 3 months long. Additionally or alternatively, according to yet another aspect, the duration is 6 months or around 6 months long.
• the historical data is the duration long portion of time-series data until essentially the time at which trend detection is being per formed.
• the present teachings can also be used to provide a monitoring system for the plant as was outlined above. Accordingly, there can also be provided a monitoring and/or control system for a plant comprising a plurality of sensors, wherein the system comprises one or more processing units configured to perform any of the method steps herein disclosed.
• a monitoring and/or control system for a plant comprising a plurality of sensors, and the system comprising one or more functionally connected processing units, the system being configured to: generate, via any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitor, via any of the one or more processing units, a level signal; wherein the level sig nal is indicative of a collective time-based variation of the time-series residual data, monitor, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generate, via any of the one or more processing units, an anomaly event signal when at a given time a value of the level signal and
• a monitoring and/or control system for a plant comprising a plurality of sensors, and the system comprising one or more functionally connected processing units, the system being configured to: provide, at any of the one or more processing units, time-series residual data of a sensor object; the sensor object being a group of at least some of the sensors from the plurality of sen sors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitor, via any of the one or more processing units, a level signal; wherein the level sig nal is indicative of a collective time-based variation of the time-series residual data, generate, via any of the one or more processing units, a level event signal; wherein the level event signal is generated when at a given time a value of the level signal changes from an expected value of the level signal at or around that time, wherein the level event signal is in
• a monitor ing and/or control system for a plant comprising a plurality of sensors, and the system comprising one or more functionally connected processing units, the system being configured to: provide, at any of the one or more processing units, time-series residual data of a sensor object; the sensor object being a group of at least some of the sensors from the plurality of sen sors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitor, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generate, via any of the one or more processing units, an association event signal; where in the association event signal is generated when at a given time a value of the level signal changes from an expected value of the association signal at or around that time, wherein the association event signal is indicative of an
• the method, or the system can be used for detecting unusual behavior and abnormal patterns in the sensor data via the proposed automated data driven technique.
• the teachings can benefit plant operator, by allowing a more efficient tracking of large sensor datasets, for example by guiding an operator’s attention at an early stage to an area of the plant where an action may be required in the future.
• the teachings can also be used to provide a pre diction of the upcoming maintenance requirements related to the plant.
• the teachings can also be used to realize an automated system for plant maintenance forecast ing and control thereof.
• the user interface can be any suitable human-machine-interface (“HMI”) that allows a user to interact with the monitoring system.
• the human machine interface (“HMI”) may comprise any one or more of; a monitoring panel, a video display unit (e.g., an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube) display, or a touch screen), an alphanumeric input device (e.g., a keyboard), a cursor control device (e.g., a mouse), and/or a signal generation device (e.g., a speaker).
• the HMI thus can, for example, be visual interface such as a panel, a screen and/or it can be an audio interface such as a loudspeaker. Accordingly, the output can either be dis played to the user and/or it can be announced via the speaker.
• the processing unit may be a computer or even a general-purpose processing device such as a microprocessor, microcontroller, central processing unit (“CPU”), or the like. More particularly, the processing unit may be a CISC (Complex Instruction Set Computing) microprocessor, RISC (Reduced Instruction Set Computing) microprocessor, VLIW (Very Long Instruction Word) mi croprocessor, or a processor implementing other instruction sets or processors implementing a combination of instruction sets.
• CISC Complex Instruction Set Computing
• RISC Reduced Instruction Set Computing
• VLIW Very Long Instruction Word
• the processing unit or processing means may also be one or more special-purpose processing devices such as an ASIC (Application-Specific Integrated Cir cuit), an FPGA (Field Programmable Gate Array), a CPLD (Complex Programmable Logic De vice), a DSP (Digital Signal Processor), a network processor, or the like.
• ASIC Application-Specific Integrated Cir cuit
• FPGA Field Programmable Gate Array
• CPLD Complex Programmable Logic De vice
• DSP Digital Signal Processor
• network processor or the like.
• the methods, systems and devices described herein may be implemented as software in a DSP, in a micro-controller, or in any other side-processor or as hardware circuit within an ASIC, CPLD, or FPGA.
• processing unit may also refer to one or more processing devices, such as a distributed system of processing devices located across multiple computer systems (e.g., cloud computing), and is not limited to a single device unless otherwise specified. Moreover, any one or more of the processing units may be located at a physical location which is different from the other processing units.
• a computer program com prising instructions which, when the instructions are executed by any one or more suitable pro cessing units of a plant monitoring and/or control system functionally connected to a plurality of sensors, cause the system to carry out the method steps herein disclosed.
• a computer program comprising instructions which, when the program is executed by one or more functionally connected processing units of a plant mon itoring and/or control system functionally connected to a plurality of sensors, cause the system to: generate, via any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitor, via any of the one or more processing units, a level signal; wherein the level sig nal is indicative of a collective time-based variation of the time-series residual data, monitor, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generate, via any of the one or more processing units, an anomaly event signal when at
• a computer pro gram comprising instructions which, when the program is executed by one or more functionally connected processing units of a plant monitoring and/or control system functionally connected to a plurality of sensors, cause the system to: provide, at any of the one or more processing units, time-series residual data of a sensor object; the sensor object being a group of at least some of the sensors from the plurality of sen sors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitor, via any of the one or more processing units, a level signal; wherein the level sig nal is indicative of a collective time-based variation of the time-series residual data, generate, via any of the one or more processing units, a level event signal; wherein the level event signal is generated when at a given time a value of the level signal changes from an expected value of the level signal at or around that
• a comput er program comprising instructions which, when the program is executed by one or more func tionally connected processing units of a plant monitoring and/or control system functionally con nected to a plurality of sensors, cause the system to: provide, at any of the one or more processing units, time-series residual data of a sensor object; the sensor object being a group of at least some of the sensors from the plurality of sen sors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitor, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generate, via any of the one or more processing units, an association event signal; where in the association event signal is generated when at a given time a value of the level signal changes from an expected value of the association signal at
• a computer-readable data carrier having stored thereon the computer program herein disclosed. Accordingly, there can also be provided a non-transitory computer readable medium storing a program causing a suitable a processing unit of a plant monitoring and/or control system to execute any method steps herein disclosed.
• a computer-readable data carrier includes any suitable data storage device on which is stored one or more sets of instructions (e.g., software) embodying any one or more of the methodolo gies or functions described herein.
• the instructions may also reside, completely or at least par tially, within the main memory and/or within the processor during execution thereof by the com puter system, main memory, and processing unit or device, which may constitute computer- readable storage media.
• the instructions may further be transmitted or received over a network via a network interface device.
• the computer program for implementing one or more of the embodiments described herein may be stored and/or distributed on a suitable medium, such as an optical storage medium or a sol id-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the internet or other wired or wireless telecommunication systems.
• a suitable medium such as an optical storage medium or a sol id-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the internet or other wired or wireless telecommunication systems.
• the computer program may also be presented over a network like the World Wide Web and can be downloaded into the working memory of a data processor from such a net work.
• a data carrier or a data storage medium for making a computer program element available for downloading can be also provided, which computer program element is arranged to perform a method according to one of the previously described embodiments.
• FIG. 1 shows an example industrial plant where certain aspects of the present teach ings can be applied
• FIG. 2 shows a block diagram with some of the signals that are generated pursuant to the present teachings
• FIG. 3 shows a plot of General Variance statistic signal pursuant to the present teachings
• FIG. 4 shows an effect of annotation on the control limits
• FIG. 5 shows certain examples of trend detection
• FIG. 1 shows an example 100 of a plant 101 for the purpose of illustrating how at least certain aspects of the present teachings can be applied.
• the plant comprises a plurality of equipment and sensors.
• the plant also comprises a processing unit 110, which is shown distributed in three parts 110a, b and c.
• a first group 102 of equipment and sensors, and a second group 103 of equipment and sensors are shown that are a part of the plant.
• the plant 101 is used for producing one or more industri al products 150.
• the product 150 can be any physical product or a service product as was out lined earlier.
• the product 150 can be a chemical or pharmaceutical product.
• the architecture or process shown in the example 100 is not of importance to the generality or scope of the present teachings.
• the first group 102 is located at a different loca tion than the second group 103.
• An intermediate product is provided by an output of the first group 102, as an input product to the second group 103.
• the intermediate product is shown provided via a transport medium such as a pipeline 188, which can, for example, be a long pipe line.
• first group 102 equipment may be rela tively isolated from the second group 103 equipment.
• first group 102 and the second group 103 There may be interdependencies between the first group 102 and the second group 103, for example, due to parameters of the intermedi ate product that is being transferred via the pipeline 188.
• Flowever there may be certain factors that are common to the two groups 102 and 103, for example, ambient pressure and tempera ture. Such ambient parameters may have an influence on the process parameters or sensor outputs on both sides. As per relevance to the process, any of such ambient parameters may thus be considered what was termed earlier as a covariant signal.
• Both the first group 102 and the second group 103 comprise a plurality of sensors, for example, temperature sensors 132, 133, 142 and 148, pressure sensors 131 , 135, 136 and 145, flow sensors 138, 143 and 147.
• the equipment in both groups includes heat exchanger 130, separa tion chamber 139, reaction tank 120, cooling unit 140, filter 151 , fan 141 , and pumps 134, 144 and 149.
• the sensors from the first group 102 are monitored by the processing means 110, or more spe cifically by a first processing unit 110a. Signals from the sensors in the first group 102 are shown received via a first communications means 105a.
• the communications means 105a can be any means, wired, wireless or their combination, that is suitable for transmitting signals or data from the sensors.
• the first communications means 105a can be a bus as shown.
• the data received by the first processing means 110a may be processed by the first processing means 110a and/or by any other processing means 110b and c. At least some data may also be stored in a memory or database 111.
• the database 111 can either be at a single place or it may be distributed as shown as 111a, b and c.
• the first pro cessing unit 110a may also perform control functions, for example, via a control bus 106a.
• the control bus 106a of the first processing unit 110a may be any communications means as dis cussed earlier in context of the bus 105a. In some cases, the bus 105a and the control bus 106a may even be the same bus or communications means.
• the control functions may include, for example, control of the pump 134.
• the processing unit 110 may even be provided by an HMI 112.
• the HMI 112 may either be provided at each of the distributed processing units 110a, b and c, as shown, or it may be provided at any one or more of them.
• the HMI may comprise a monitoring panel or a video screen and one or more input devices such as a keyboard or mouse for a user to interact with the processing means 110.
• the HMI may also comprise an audio de vice such as a loudspeaker. Events such as alarms may be communicated audibly and/or visu ally via the HMI.
• the sensors from the second group 103 are monitored by the processing means 110, or more specifically by a second processing means 110b.
• Signals from the sensors in the sec ond group 103 are shown received via a second communications means 105b.
• the second communications means 105b can be any means, wired, wireless or their combination, that is suitable for transmitting signals or data from the sensors.
• the second communica tions means 105b can be a bus as shown.
• the data received by the second processing means 110b may be processed by the second processing means 110b and/or by any other processing means 110a and c.
• at least some data may also be stored in a memory or database 111.
• the second processing unit 110b may also perform control func tions, for example, via a second control bus 106b.
• the control functions may include, for exam ple, control of: the pumps 144 and 149, the fan 141, and valve 146.
• the second control bus 106b of the second processing unit 110b may be any communications means. In some cases, the bus 105b and the control bus 106b may even be the same bus or communications means.
• the first processing unit 110a and the second processing unit 110b are functionally connected via data link 190, which may be any suitable communications medium, wired, wireless or their combination.
• the processing units are thus able to exchange data that may include any data or signals such as sensor data, status data and event signals.
• the data link may even be used for transferring data from one database or memory to the other.
• a separate processing unit e.g., a third processing unit 110c
• the third processing unit 110c may be at a higher hierarchy and may be a plant level monitoring and/or control system.
• the third processing unit 110c may either be at the same location as the plant, or it may even at least partially be at another location than the plant, for example, it may be a cloud-based platform.
• the third processing unit 110c may be within the plant, but its database 111c may be implemented as a cloud storage or vice versa.
• the supervi- sory processing unit 110c may even be located at another plant located at a different site than the plant 101.
• the third processing unit 110c may even be located in between the first and the second processing units 110a and b, i.e., data link 190 being divided into two sections, first between the first unit 110a and the third unit 110c, and the second between the third unit 110c and the second unit 110b.
• data link 190 being divided into two sections, first between the first unit 110a and the third unit 110c, and the second between the third unit 110c and the second unit 110b.
• a specific architecture of the processing units or the plant is not essential to the scope or generality of the present teachings.
• the another plant may even be located in another country.
• the first group 102 and the second group 103 are located in different plants or countries.
• a supplier plant and a consumer plant connected via a gas pipeline may be located in different countries.
• processing units 110a, b and c, and the databases 111 a, b, c may be implemented as a cloud-based service, for example provided by a third-party.
• the processing units 110a, b and c, and/or the databases 111a, b, c may be at the same place or they may even be the same unit.
• a conventional system may monitor the state of one or more sensors individually. For example, the output signal from the temperature sensor 148. A rising temperature may be used to indicate overheating of the pump 149, for example due to a reumbled flow caused by a blockage in the filter 151. Flowever, in reality, the increase in the tem perature may have been caused due to ambient temperature increase. Such a system may thus lead to false positive events indicating an anomaly.
• the second processing unit 111b may compare the measured or observed output value of the temperature sensor 148 from its expected value at that time.
• the expected value may be generated by an expected state model of the temperature sensor 148.
• the model e.g., neural network may be trained using the historical time series data of the sensor 148, preferably under desired operating condi tions.
• the expected state model may be input with covariate signals that influence the output of the sensor 148.
• ambient temperature may be one of the covariate signals.
• the processing means 110 may use the entire co variate pool of the whole plant 101 to check, using the historical data, which of the covariates have an effect, or possess predictive power, on the output of the sensor 148.
• the covariates that have measurable influence on the output of the sensor 148 are thus selected as model in puts.
• residual signal is generated for the output of the sensor 148, which is a difference between the observed sensor output value and the output of the expected state model at that given time. If the sensor is behaving normal ly, the residual signal will be mostly random noise.
• the sensor ob ject refers to a group of sensor residual signals that are consolidated and monitored together.
• the group of sensor residual signals are time series residual signals that are received from a pre-selected plurality of sensors.
• the pre-selected plurality of sensors may either be selected manually, or at least partially automatically via the processing unit 110.
• the processing unit may decide this for example, based on similarity in sensor response, sensor types, covariate de pendencies, or their combination.
• the group of sensor residual signals or residual data are then analyzed by the processing unit 110 to compute a level signal.
• the level signal is indicative of a collective time-based variation of the time-series residual data.
• the time dependent level signal value is then compared with an expected level signal value at that time.
• the processing unit 110 may generate a level event signal at any given time when the value of the level signal changes beyond the expected level signal value at or around that time.
• the level event signal is deemed indicative of an anomaly in at least one of the equipment in the plant.
• the processing unit 110 may issue an alarm. Furthermore, the processing unit 110 may check for which of the sensors in the sensor object the sensor output breached the expected sensor output value at or around the time the level event signal was generated. This is used by the processing unit 110 to find the source of the anomaly.
• the expected level signal value is preferably generated by the processing unit 110 using a sen sor object model.
• the sensor object model is a predictive model or a neural network that has been trained using historical residual data.
• the expected level signal is provided as a value range within which the level signal may lie.
• one or more limit values for the level signal may be provided.
• the level event signal is generated when the observed level signal value goes beyond an expected level signal limit.
• the expected level signal limit may be an upper expected level signal limit and/or a lower expected level signal limit.
• the processing unit computes another score for making the anomaly detection fur ther immune to noisy spikes in the residual signal.
• a time-dependent association signal value is generated.
• the group of sensor residual signals or residual data are thus analyzed by the processing unit 110 to compute an association signal.
• the association signal is indicative of the variation and/or association structure of the time-series residual data.
• the time dependent association signal value is then compared with an expected association signal value at that time.
• the processing unit 110 may generate an association event signal at any given time when the value of the association signal changes beyond the expected association signal value at or around that time.
• the association event signal is deemed indicative of an anomaly in at least one of the equipment in the plant.
• the processing unit 110 may issue an alarm.
• the processing unit 110 may check for which of the sensors in the sensor object the sensor output breached the expected sensor output value at or around the time the level event signal was generated. This may also be used by the processing unit 110 to find the source of the anomaly.
• the expected association signal value is preferably generated by the processing unit 110 using the sensor object model.
• the expected association signal is provided as a value range within which the asso ciation signal may lie.
• one or more limit values for the level signal may be provided.
• the association event signal is generated when the observed association signal value goes beyond an expected association signal limit.
• the expected association signal limit may be an upper ex pected association signal limit and/or a lower expected association signal limit.
• the limits can also be termed as control limits.
• Breaching of any one or both of the level signal and association signal their respective expected or limit values may be considered as indicative of an anomaly.
• the processing unit 110 may even perform a trend detection. Due to retraining of the models, slow moving drifts may get eliminated from observation by the level and association monitoring.
• the processing unit 110 may specific functions may have been referred to as being performed by “the processing unit 110”, it will be understood that in some cases it may even be implemented as being performed via any of the one or more processing units 110a, b and c. It will also be understood that in some cases there may be additional processing units. For exam ple, some sensors may even be provided a dedicated processor that is configured to calculate the residual signal for that sensor. In that case, the residual signal for such sensors may be di rectly provided at the processing unit 110 as an input.
• the processing unit 110 may monitor another sensor object via the corresponding one or both level and association signals for that object.
• Each group may have more than one sensor ob ject.
• the processing unit may backtrack the sensor data to find the source of anomaly, for example as outlined previously. Additionally, the processing unit may forecast the maintenance requirements for the anomaly, for example by providing an estimated date or time by which maintenance should be performed to prevent a certain disruption. Disrup tion may be calculated as a loss of productivity or as wastage as compared to a planned shut down for maintenance.
• FIG. 2 shows a block diagram 200 representing the signals that are generated and monitored. Charts on the left side show observed and expected values of five different sensors that are grouped in a sensor object. The grouping is preferably done automatically, for example, using self-organizing maps, but it may even be done at least partially based on manual feedback.
• First set of curves 201a pertains to measured output signal from a first sensor and the expected output thereof.
• curves 201b - e pertain each to the measured output signal from a second to fifth sensor respectively and their expected outputs.
• respective residual signals 202a - e are obtained.
• the first residual signal 202a pertains to the first sensor.
• the residual signals 202a - e are more homo geneous. As discussed previously, superfluous information from the sensor outputs can been removed as a virtue of generating the residual signals.
• the signals are time-dependent, or they comprise time-series values.
• a sensor object 204 is realized that comprises multidi mensional residual data 203.
• time-dependent level signal or score 205 is shown generated.
• the level signal 205 is provided an expected level signal limit value 207, which represents a probability space of expected values within which the level signal may validly lie.
• the expected level signal limit value 207 may also be a time-dependent value. As shown, just after time 209, the expected level signal limit value 207 is reduced by the sensor object model.
• 205p represents a peak in the level signal 205 when said signal changed beyond an expected value of the level signal, or an expected level signal limit value 207 that that time. Accordingly, in such a case a level event signal would be generated by the processing unit 110.
• the processing unit may then trace the root cause of the anomaly, for example, by analyzing one or more of the sensor signals 201 a - e.
• the processing unit may use effect size calculation for finding one or more sensors that cause the most contribution to the signal change.
• An alarm may be displayed on a visual monitoring panel 210. For example, the relevant equipment on the panel may be highlighted.
• time-dependent association signal or score 206 is shown gen erated.
• the association signal 206 is provided an expected association signal limit value 208, or more specifically, an upper association signal limit value 208a and a lower association signal limit value 208b.
• the distance between these limits represents a probability space of expected values within which the association signal may validly lie.
• the expected association signal limit values 208a and b may also be time-dependent values. It can be seen that 206p represents a peak portion in the association signal 206 when said signal changed beyond upper expected values of the association signal, or an upper expected association signal limit value 208a that that time. Accordingly, in such a case an association event signal would be generated by the processing unit 110.
• the processing unit may then trace the cause of the anomaly, for example, by analyzing one or more of the sensor signals 201 a - e.
• the processing unit may use effect size calculation for finding one or more sensors that cause the most contribution to the signal change.
• the association score can also detect the rate of change of the movement of the resid ual signals. Similarly, an alarm may be displayed on the visual monitoring panel 210.
• the event signal for either or both scores is caused by an activ ity in the plant which results in a deviation in the residual data from the expected states.
• Such an activity may be a repair or other event that changes be behavior of the observed states.
• the user may be aware what the event is caused by. The user then has a possibility to annotate the event according to a specific classification or type.
• One or more annotations can thus be fed back to the model such that the model is trained to classify such events in the fu ture.
• FIG. 3 shows a plot 300 of Generalized Variance (“GV”) that can be used for computing the var iation and association structure of the sensor object data. GV can thus be used to generate the association signal 206.
• the plot 300 shows an association signal 306 with Generalized Vari ance value on the Y-axis 301 and time on the X-axis 302.
• Limit values 308 are also shown which may also be called an upper control limit (“UCL”) 308a and a lower control limit (“LCL”) 308b.
• the distance 310 between these values 308a and b represents a control range or a prob ability space within which the value of the association signal 306 may validly lie.
• the values lying within the control range 310 at any given time may thus be called expected values at that time.
• the control range 310 may also be time dependent, however in this case, it is shown con stant.
• association signal 306 changed from an expected value, or in this case changed beyond the lower control limit 308b at or around that time. Accordingly, in this case an association event signal would be created or gen erated.
• association signal 306 changed again from an expected value, or in this case changed first beyond the lower control limit 308b and then beyond the upper control limit 308a. Accordingly, in this case also one or more association event signals would be generated.
• Such a chart 300 may also be called a control chart that is used for monitoring numerical statis tics of the signals over time.
• a similar control chart may also be generated for the level signal value statistic.
• FIG. 4 shows a couple of charts 400 illustrating how annotation may be used to improve the focus or relevance of training data.
• the charts 400 are shown as control charts.
• the left control chart 430a shows the control limits 410 and 420a for a score signal 490.
• the score signal in this case is an association signal.
• Y-axis is an association signal value, e.g., GV statis tic.
• the X-axis 302 represents time.
• the charts have a peak region 455 which shows abruptly high values for the signal 490. Such high values may have occurred due to an abnormal event such as a maintenance activity in a certain portion of the equipment that includes sensors within the sensor object.
• the charts also show a start time 401 and an end time 402 compose a time window that can be used for training the sensor model.
• the control limits would be determined as shown, i.e., an upper control limit 420a for the unannotated chart 430a and a lower control limit 410. It will be appreciated that such a high upper control limit 420a may not be appropriate to detect a change in the score sig nal from an expected value at that time. Accordingly, some anomaly events may not be flagged.
• choice of time window i.e., time-period enclosed between the start time 401 and the end time 402 is also selected by the processing means, such that the selected win dow reflects the normal behavior of the sensor object.
• Annotations may also be used to mark such windows that are desirable. The control limits can thus be adjusted such that anomalies can be better detected.
• control limits may be statistical quantile limits.
• FIG. 5 shows example charts 501 - 500 for demonstrating how the processing means may per form trend detection.
• Trend detection as proposed comprises computing three metric values on long term data within time windows at least 1 week long. The three metric values are Strength, Smoothness, and Currentness or Actuality. Trend detection is also used on sensor object.
• a value between 0 and 1 is assigned for each of the metrics, where 0 indi cates false and 1 indicates true. Usually the values are between 0 and 1 , indicating a probability of the metric or property of the trend.
• the values indicated below are examples and the limits specified above are not absolute.
• a scaling factor may be applied, such as any of the value lies between 0 and 100. The values mentioned herein thus should not be taken specified in an absolute sense.
• For specifying the direction of the trend a sign may also be assigned to the strength value.
• the first chart 501 shows a signal in rising trend
• the metric values computed for the first chart 501 are: Strength: 1 , Smoothness: 0.9, Actuality: 1.
• the second chart 502 shows an abruptly rising signal encompassed between a lower noise floor and an upper noise floor. The trend has thus almost died out.
• the metric values computed for the second chart 502 are: Strength: 0.69, Smoothness: 0.98, Actuality: 0.25.
• the third chart 503 shows a signal in the form of a triangular waveform. The metric values com puted for the third chart 503 are: Strength: 0.1, Smoothness: 0, Actuality: 0.24.
• the fourth chart 504 shows a signal with an initial noisy portion and then a rising trend.
• the metric values computed for the fourth chart 504 are: Strength: 0.73, Smoothness: 0.97, Actuali ty: 1.
• the fifth chart 505 shows a signal with an initial rising trend and then an upper plateau.
• the metric values computed for the fourth chart 504 are: Strength: 0.77, Smoothness: 0.97, Actuali ty: 0.28.
• a method for monitoring a plant comprising a plurality of sensors, and one or more functionally connected processing units, the method comprising: providing, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residue signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitoring, via any of the one or more processing units, a level signal; wherein the level signal is indicative of a collective time-based variation of the time-series residual data, monitoring, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generating, via any of the one or more processing units, an anomaly event signal when at a given time a value of the level signal and/or a value of the association signal changes from an expected value of the respective signal at
• any of the respective expected value is provided as a corresponding expected value limit specifying for a given time a plurality of expected val ues, as a range and/or as discrete values, that the corresponding signal may validly have with out the anomaly event being generated.
• the method also comprises: determining, in response to the anomaly event signal, at least one root cause of the anomaly, by performing any one or more of the: checking for which of the sensors in the sensor object the sensor’s measured output changed from the sensor’s expected output at or around the same time as the occurrence of the anomaly event; analyzing the time series residual signal of each sensor within the sensor object to determine one or more main drivers or most domi nant contributors to the level signal value; analyzing the time series residual signal of each sen sor within the sensor object to determine one or more main drivers or most dominant contribu tors to the association signal value; and analyzing covariance of time series residual signals of each pair combination of the sensor residual signals within the sensor object to determine one or more main drivers or most dominant contributors to the association signal value.
• the method also comprises: determining, in response to the anomaly event signal, state of health of at least one equipment related to the sensor object.
• any of the expected value or the expected limit value is provided by a sensor object model that is a predictive model which has been trained using historical residual data of the sensor object.
• each covariate signal being a signal representing a parameter upon which at least one of the residual signals are dependent upon.
• each covariate signal being a signal representing a parameter up on which the sensor’s output is dependent upon.
• Clause 10 The method according to any of the above clauses, wherein the sensor object is provided by at least partially automatically grouping the at least some of the sensors, for example, using at least one data-centric algorithm, such as a clustering algorithm, for example a self-organizing map algorithm, further for example, the sensor object being at least partially automatically gen erated by any of the one or more processing units using at least one self-organizing map.
• at least one data-centric algorithm such as a clustering algorithm, for example a self-organizing map algorithm
• the expected state model is select ed automatically by the processing unit by analyzing a plurality of different predictive model types, and selecting the model type as the expected state model which provides the lowest er ror between: the output of that model when trained with a specific training window of the histori cal time-series data; and the actual historical sensor output within a specific time-window of the historical time-series data.
• the level signal value is generated using a distance estimator indicating the time at which and the amount by which the time-series residual data deviates from its normal or expected or mean state.
• association signal value is gen erated using a statistical measure of multivariate dependencies in the residual data, or to meas ure at a given time the dispersion of the time-series residual data.
• method also comprises: detecting, via any of the one or more processing units, a drift in the output signal a sensor; wherein the sensor is among the at least some of the sensors, and wherein the drift is comput ed from historical time series data of the sensor, the historical data of the sensor being from a time period that is at least 1 week long, and wherein the drift is detected by computing the strength, smoothness and currentness of the historical data of the sensor.
• a method for monitoring a plant comprising a plurality of sensors, and one or more functionally connected processing units, the method comprising: providing, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitoring, via any of the one or more processing units, a level signal; wherein the level signal is indicative of a collective time-based variation of the time-series residual data, generating, via any of the one or more processing units, a level event signal; wherein the level event signal is generated when at a given time a value of the level signal changes from an expected value of the level signal at or around that time, wherein the level event signal is indica tive of an anomaly in at least one of the equipment in the plant.
• a method for monitoring a plant comprising a plurality of sensors, and one or more functionally connected processing units, the method comprising: providing, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residual signal which is a difference between the sensor’s measured output and the sensor’s expected output, monitoring, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generating, via any of the one or more processing units, an association event signal; wherein the association event signal is generated when at a given time a value of the level sig nal changes from an expected value of the association signal at or around that time, wherein the association event signal is indicative of an anomaly in at least one of the equipment in the plant.
• a method for monitoring a plant comprising a plurality of sensors, and one or more functionally connected processing units, the method comprising: providing, at any of the one or more processing units, time-series residual data of a sen sor object; the sensor object being a group of at least some of the sensors from the plurality of sensors, and wherein the residual data comprises, for each of the sensors of the sensor object, a residue signal which is a difference between the sensor’s measured output and the sensor’s expected output, wherein the sensor object is provided by at least partially automatically group ing the at least some of the sensors, monitoring, via any of the one or more processing units, a level signal; wherein the level signal is indicative of a collective time-based variation of the time-series residual data, monitoring, via any of the one or more processing units, an association signal; wherein the association signal is indicative of the variation and/or association structure of the time-series residual data, generating, via any of the one or more processing units, an anomaly event signal when at a given time a value of
• a monitoring and/or control system for a plant comprising a plurality of sensors, wherein the system comprises one or more processing units configured to perform the method steps of any of the clauses 1 - 17. Clause 19.
• a computer program comprising instructions which, when executed by a processing unit of a plant monitoring and/or control system functionally connected to a plurality of sensors, cause the system to carry out the method steps of any of the clauses 1 - 17.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Automation & Control Theory (AREA)
• Mathematical Physics (AREA)
• Evolutionary Computation (AREA)
• Artificial Intelligence (AREA)
• Mathematical Analysis (AREA)
• Mathematical Optimization (AREA)
• Probability & Statistics with Applications (AREA)
• Pure & Applied Mathematics (AREA)
• Algebra (AREA)
• General Engineering & Computer Science (AREA)
• Manufacturing & Machinery (AREA)
• Quality & Reliability (AREA)
• Testing And Monitoring For Control Systems (AREA)
• Testing Or Calibration Of Command Recording Devices (AREA)

Priority Applications (5)

Applications Claiming Priority (2)

Publications (1)

Family

ID=69468491

Family Applications (1)

Country Status (6)

Cited By (4)

Families Citing this family (14)

Citations (2)

Family Cites Families (7)

• 2021
  • 2021-01-29 EP EP21702960.2A patent/EP4100808A1/en active Pending
  • 2021-01-29 JP JP2022547299A patent/JP7680458B2/ja active Active
  • 2021-01-29 WO PCT/EP2021/052193 patent/WO2021156157A1/en not_active Ceased
  • 2021-01-29 KR KR1020227030188A patent/KR20220137931A/ko active Pending
  • 2021-01-29 CN CN202180012332.XA patent/CN115039047A/zh active Pending
  • 2021-01-29 US US17/796,698 patent/US11971706B2/en active Active

Patent Citations (2)

Non-Patent Citations (3)

Also Published As

Similar Documents

Legal Events