Decision Support in Industrial Buildings

Abstract

Embodiments of the present disclosure relate to a decision support system and method in an industrial plant. A decision support system and method for an industrial plant are provided. The decision support system is configured to: obtain a causal diagram that models a causal hypothesis related to conditional dependencies between variables in the industrial plant; obtain observational data related to the operation of the industrial plant; and perform causal inference using the causal diagram and the observational data to estimate at least one causal effect related to making a decision during the operation of the industrial plant.

Description

Technical Field

[0001] The present invention relates to decision support in industrial plants. Background Art

[0002] Industrial plants are large and complex scenarios. Process control systems provide a means for operators to monitor and control processes, for example by providing visualization of process and alarm management information, while process historians store the state of the system over time. Process visualization typically describes the various parts of a process in a hierarchical manner, showing dynamic process variables, trends, and alarm states for example of process sections or specific operator tasks. These process visualizations are based on engineering data and process models and reflect for example the flow of information, materials, or energy in the process. However, compared to the information contained in engineering artifacts, the process topology information reflected in process visualizations is typically incomplete and fragmented.

[0003] Monitoring the process allows operators to diagnose abnormal situations and take corrective actions. To reduce production losses, operators must quickly evaluate the possible impact of corrective actions. However, the capabilities of operators are related to their experience. Gaining experience in operating a specific plant typically takes several years. For inexperienced operators, it can be challenging to find the required information in a timely manner. Clearly, there are no experienced operators in newly built plants. Even in brownfield plants, experienced operators leave and need to be replaced. Operator Training Simulators (OTS) can be used to provide an advantage in operations, but are costly due to the need for high-fidelity process models. Instead, operations that are limited in efficiency in the early stages of operator training are typically accepted.

[0004] For these reasons, the owners of industrial plants face the risk of inefficient plant operation and even major accidents due to inexperienced operators. Summary of the Invention

[0005] Therefore, there is a need for an affordable solution that can support inexperienced operators in making decisions related to the operation of industrial plants. The subject matter of the independent claims meets this need. The dependent claims set forth optional features.

[0006] According to a first aspect, there is provided a decision support system for an industrial plant, the decision support system being configured to: obtain a causal graph that models a causal hypothesis related to a conditional correlation between variables in the industrial plant; obtain observation data related to the operation of the industrial plant; and perform causal inference using the causal graph and the observation data to estimate at least one causal effect related to making a decision during the operation of the industrial plant.

[0007] In one example, the causal graph is a directed acyclic graph, but it should be understood that any probabilistic graphical model can be used to represent conditional dependencies between variables. The directed acyclic graph includes a set of vertices or nodes representing variables in an industrial plant and directed edges or arrows, each edge interconnecting two vertices. Each vertex represents an independent or dependent variable. A "causal hypothesis" refers to the assumption of a causal relationship between two variables, but the causal relationship (has not) been verified by observational data. Thus, this document refers to possible (i.e., hypothesized or presumed) causal relationships rather than inferred causal relationships verified by observational data. Causal hypotheses are based on domain knowledge, such as engineering data related to an industrial plant. The causal graph can be constructed manually, for example, by a subject matter expert (SME), or automatically, for example, by a modeling tool such as described herein. One or more causal graphs can be used to capture all dependencies between variables in an industrial plant.

[0008] Observational data relates to previous operations of the industrial plant and can include collected measurement data or historical measurement data (e.g., from operations and / or commissioning), such as time series data related to one or more (preferably all) of the variables represented by the causal graph. Observational data can thus be considered non-experimental or empirical data. Although experimental data obtained using interventions or high-fidelity simulations can also be used, it can be understood that this is not always possible or typically too expensive.

[0009] In one example, performing causal inference includes: identifying a causal effect based on an input query; estimating the causal effect; optionally verifying the causal effect; and presenting the causal effect.

[0010] Identifying a causal effect can include finding all possible paths between two variables in the causal graph. For example, the input query (or estimand) can specify a causal relationship or equivalence between two variables, for example, in the form of "Does variable B affect variable A?". In other words, the query designates one variable as the action, one variable as the outcome, and optionally one or more other variables as, for example, confounders. All possible ways in which B has a causal effect on A are then identified. Invalid paths, such as backdoor paths, can be excluded from consideration.

[0011] Estimating the causal effect can include using conditional probability formulas to estimate the strength of the causal relationship between two variables. For example, for two variables A and B in the input query, the formula can include:

[0012] P1 = P(A|B, confounders) and P2 = P(A|confounders),

[0013] Among them, the confounding factor is a variable (measured or controlled) in the causal diagram that affects A and B. The strength is represented by the difference between P1 and P2. The greater the difference between P1 and P2, the stronger the causal relationship. The decision support system can be configured to use methods to estimate the statistical significance of the difference between P1 and P2, including, for example, one or more of propensity-based stratification, propensity score matching, inverse propensity weighting, regression, regression discontinuity. Further estimation methods will be obvious to those skilled in the art. Additionally or alternatively, estimating the causal effect can include determining that there is no causal relationship between two variables. For example, in response to determining that P1 is equal to P2 (e.g., if the difference does not exceed a predetermined threshold), no causal relationship between A and B can be inferred (in other words, each statistical relationship is explained by the confounding factor).

[0014] Verifying the causal effect can include: performing data subset verification by re-estimating the strength of the causal relationship using only a subset of the observational data (and verifying the causal effect in response to no significant change in the causal effect) and / or performing a control treatment (placebo treatment) by re-estimating the strength of the causal relationship using the observational data, where the data including B is replaced with data where B does not exist (and verifying the causal effect when the estimated strength drops to zero or approximately zero). Additional verification methods will be obvious to those skilled in the art.

[0015] Presenting the causal effect can include, for example, outputting the causal effect to a human plant operator via the HMI or dashboard of a process control system. Additionally or alternatively, presenting the causal effect can include outputting the causal effect to an entity that performs automated or semi-automated operations of an industrial plant, such as an automated plant operating system, e.g., a plant operation agent based on reinforcement learning.

[0016] The decision support system can be configured to perform root cause analysis related to a given (measured) variable in an industrial plant to identify which other variables in the causal diagram are most likely to affect the given variable. Root cause analysis can be performed for diagnostic purposes when a given variable is found to be abnormal or exhibits an undesirable behavior. For example, when an anomaly is found among several measured variables, root cause analysis can be repeated for some or all of the measured variables in the industrial plant.

[0017] In one example, performing root cause analysis on a given variable includes: performing causal inference to estimate the strength of the causal relationship between the given variable and each other variable. For example, in an industrial plant operating using process variables PV1 - PV n and the given variable is PV iIn the case of , performing root cause analysis may include performing causal inference as described above using multiple input queries, each input query designating a given variable as an outcome variable and one of the other variables as an action variable. More specifically, for the case where the given variable A=PV is not included i Other variables B = PV1...PV n In each of the causal graphs, the input query may be in the form of "Does variable B affect variable A?" or equivalent. The other variables may be ranked according to the strength of the causal relationship with the effect of the given variable. Identifying which other variables in the causal graph of the given variable are most likely to affect the given variable may include identifying the other variables associated with the strongest causal relationship as the most likely root cause, or alternatively returning a list of other variables ranked by their respective causal relationship strengths with the given variable. Variables that have no or only a weak causal relationship with the given variable may be excluded.

[0018] In other words, performing a root cause analysis may include defining a problem based on the values of a given variable that satisfy certain conditions, and performing statistical estimation to calculate, for each of the other variables, the probability of the problem occurring given the observed data for the other variables. For example, performing a root cause analysis may include performing a root cause analysis for all of the variables that do not include a given variable A=PV i Each other variable B = PV1...PV n , evaluates P(A=a|do(B=b)), where "a" represents a problem condition in a given variable A. The operator "do" is known in the art for simulating an intervention by reducing the observed data to only that portion containing the specified value. Other variables can then be identified as possible root causes of the problem and / or ranked based on the corresponding calculated probabilities. The condition can be, for example, a value below or above a predetermined threshold or outside a predetermined range, such as a temperature exceeding a threshold to be classified as a hotspot.

[0019] The decision support system may be configured to find corrective actions that can affect changes in a given measured process variable, for example by using cause-effect diagrams and observational data to identify controlled variables in an industrial plant that exhibit a cause-effect relationship with a given variable that should be corrected.

[0020] In one example, finding the corrective action includes performing causal inference to estimate the strength of the causal relationship between a given variable and each other variable in the causal graph that is a controlled variable. For example, in an industrial plant using process variables PV1-PV2 including at least one controlled variable C nIn the case of performing an operation, finding a corrective action may include, for each controlled variable, performing causal inference as described above using an input query that specifies a given variable as the outcome variable and a controlled variable as the action variable in the form of, for example, "Does variable C affect variable A?" or an equivalent form, where C is the controlled variable C, where C != A, and where A = PVi. The controlled variables may be ranked according to the strength of their respective causal relationships with the given variable. The controlled variable associated with the strongest causal relationship may be identified for the corrective action, or alternatively, a list of controlled variables may optionally be ranked according to the strength of their respective causal relationships with the given variable. Variables that have no or only a weak causal relationship with the given variable may be excluded. Preferably, the decision support system identifies as few corrective actions as possible that have the desired effect. The decision support system may be configured to use a heuristic algorithm, such as a rule-based algorithm, to select, exclude, or prefer corrective actions, for example, preferring a setpoint change over directly manipulating an actuator. As described below, a what-if analysis may also be performed to eliminate corrective actions.

[0021] The decision support system may be configured to perform a what-if analysis to identify one or more possible side effects of changing a controlled variable in an industrial plant. More specifically, performing a what-if analysis includes: identifying a first variable A that is affected by both a second and a third variable B and C that do not affect each other in a causal graph, and performing causal inference to estimate the extent to which the causal effect of the second variable B on the first variable A is modified by the third variable C. For example, performing a what-if analysis may include, for all values b of B, evaluating P(A|do(B = b), C = c), while c remains constant. Here, the assumed query is: What if the controlled variable B is set to b while the value of another variable C is c? A is the first (cause) variable, B is the second variable, and C is the third variable. The second variable B may be the controlled variable. In this way, the side effect of changing the controlled variable can be determined, i.e., the effect of the second variable B on the first variable A when the value of the third variable C is c. This process may be repeated for each causal graph in which a controlled variable is represented to list all possible side effects of modifying the controlled variable. For example, for each graph in which there is a controlled variable, the input query may relate to whether the controlled variable has a causal effect on the (multiple) outcome variables of the causal graph. The results for all causal graphs constitute a list of possible side effects.

[0022] A decision support system can be configured to optionally perform a combination of root cause analysis, corrective action search, and what-if analysis in a sequential manner. In one example, the decision support system is configured to sequentially perform: 1) root cause analysis to identify one or more candidate root causes of a process anomaly; 2) corrective action search to identify controlled variables that can resolve the process anomaly; 3) what-if analysis to explain what may occur when the control variables are changed, i.e., analyze the possible side effects of the corrective action. Corrective actions with fewer side effects may be preferred. The resulting corrective actions can be output to the operator and / or an automated plant operating system.

[0023] According to a second aspect, there is provided a decision support method for an industrial plant, the decision support method comprising: obtaining a causal graph that models a causal hypothesis related to the conditional dependencies between variables in the industrial plant; receiving observational data related to the operation of the industrial plant; and performing causal inference using the causal graph and the observational data to estimate at least one causal effect related to making decisions during the operation of the industrial plant.

[0024] The decision support system and method of the first and second aspects thus improve plant efficiency in the case of inexperienced operators, reduce the dependence on the cognitive state of the operator, and provide more effective handling of complex situations and faster resolution of operational problems. Regardless of the physical distances between elements in the plant topology, the operator can help draw conclusions about how problems are connected.

[0025] Searching for possible root causes of a problem helps the operator evaluate possible mitigation measures, i.e., infer the consequences of actions taken by the operator to solve the problem. The decision support system helps the operator better analyze the problem and find the root cause faster. Thus, less time is required to understand the root cause of the problem, enabling the problem to be solved quickly and correctly. The decision support system helps the operator identify the root cause of problems related to temporal and topological dependencies that may occur in the industrial plant, such as valve anomalies.

[0026] By effectively and efficiently solving the problem of anomalies, searching for corrective actions can improve the reliability of the plant.

[0027] What-if analysis enables the operator to make faster and more informed decisions regarding corrective actions by explaining the context and predicted consequences of the corrective action, reducing adverse side effects.

[0028] In addition, the decision support system facilitates autonomous plant operation using artificial intelligence or other automated plant operating systems by identifying the root cause of problems and effectively selecting corrective actions.

[0029] As described above, a causal diagram (or any probabilistic diagram) can be modeled using domain knowledge, including engineering data such as plant topology data, process model data, and control specifications.

[0030] According to a third aspect, there is thus provided a modeling tool for constructing a causal diagram that models causal hypotheses related to conditional dependencies between variables in an industrial plant. The modeling tool is configured to: receive engineering data related to an industrial plant; use the engineering data to identify a plurality of variables in the industrial plant; use the engineering data to identify connections between the variables; and construct a causal diagram that includes a set of nodes representing the identified variables and a set of directed edges interconnecting the nodes, the directed edges representing causal hypotheses related to conditional dependencies between the variables based on the identified connections.

[0031] The engineering data can include process topology data, which can include a graph containing topological elements (instruments, control loops, pipes, vessels, etc.) and their connections (material, information, energy flows). In the case where a topology graph is available, the modeling tool can be configured to extract nodes and / or edges from the topology graph using graph analysis algorithms or heuristic algorithms. The engineering data can also include one or more of the following: i) naming conventions, e.g., tag names; ii) information available in process control application configurations, such as connections between process variables, I / O points, and properties, e.g., OPC DA; iii) information available in process history configurations. The engineering data can include information from process engineering, control application engineering (including control application library construction), and control systems. The engineering data can include P&IDs. The engineering data can further include a control narrative providing a text description of control logic on the basis of which a causal diagram can be constructed.

[0032] The variables can be identified using, for example, one or more of the following: i) process control system state information in the engineering data, such as a set of process variables or alarm conditions; ii) process topology data.

[0033] The connections can be identified, for example, based on the material, energy, and information flows identified in the engineering data. The connections may involve connections between elements in the process topology and / or connections between control system objects, e.g., control modules, I / O objects / labels, aspect objects, alarm conditions, event history measurements, and events / time series, and objects in the topology model.

[0034] Constructing the diagram can also include using one or more propensity-based techniques and / or covariate matching techniques. Constructing the diagram can include using a graph construction algorithm. The algorithm can use a reference model that optionally defines the types of nodes and edges and the hierarchical relationships of the node types.

[0035] According to a fourth aspect, there is provided a modeling method for constructing a causal graph, which models causal hypotheses related to conditional correlations between variables in an industrial plant. The method includes: receiving engineering data related to the industrial plant; using the engineering data to identify a plurality of variables in the industrial plant; using the engineering data to identify connections between the variables; and constructing a causal graph, which includes a set of nodes representing the identified variables and a set of directed edges interconnecting the nodes, where the directed edges represent causal hypotheses related to conditional correlations between the variables based on the identified connections.

[0036] Modeling the industrial plant according to the third and fourth aspects helps to avoid cumbersome manual selection and avoid preparing inputs from various information sources to analyze (abnormal) plant situations.

[0037] According to a fifth aspect, there is provided a computing device, which includes a processor configured to execute the method according to the second or fourth aspect.

[0038] According to a sixth aspect, there is provided a computer program product including instructions, which, when executed by a computing device, cause the computing device to execute the method according to the second or fourth aspect.

[0039] According to a seventh aspect, there is provided a computer-readable medium including instructions, which, when executed by a computing device, cause the computing device to execute the method according to the second or fourth aspect.

[0040] Whether specifically disclosed in this combination or separately, the present invention may include one or more aspects, examples, or features, either separately or in combination. Any optional feature or sub-aspect of one of the above aspects is applicable to any other aspect.

[0041] These and other aspects of the present invention will become apparent from the embodiments described hereinafter and will be elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] A detailed description will now be given by way of example with reference to the accompanying drawings, in which:

[0043] Figure 1 illustrates an industrial plant according to the present disclosure;

[0044] Figure 2 is a flowchart showing a method of performing causal inference;

[0045] Figure 3 is a topological graph;

[0046] Figure 4 is a causal graph;

[0047] Figure 5 illustrates the modeling method; and

[0048] Figure 6 Illustrated is a computing device that can be used according to the systems and methods disclosed herein. Detailed Description

[0049] Figure 1 Schematically illustrated is an industrial plant 100 that executes an industrial process 102. The process 102 is affected by process equipment 104 (such as an actuator), and the process equipment 104 receives control signals CPV1, CPV2... CPV that specify a plurality of controlled variables (such as actuator values or setpoints) as inputs. i The sensors 106 provide feedback on the process 102 in the form of a plurality of measured variables MPV1, MPV2... MPV j . The controlled variables and the measured variables together can be referred to as process variables PV (PV1, PV2... PV n ). The process control system 108 receives the feedback from the sensors 106 and outputs control signals to the process equipment 104 so as to affect the process 102 in a manner known in the art to produce the desired product.

[0050] According to the present disclosure, the industrial plant 100 further includes a decision support system 110 that is configured to: obtain a causal graph 118 that models causal hypotheses regarding conditional correlations between variables in the industrial plant 100; obtain observation data 112 related to the operation of the industrial plant 100; and perform causal inference using the causal graph 118 and the observation data 112 to estimate at least one causal effect 120 related to making decisions when operating the industrial plant 100.

[0051] Although only one causal graph 118 is illustrated and described, several causal graphs can be used to represent the industrial process 102. The causal graph 118 is i) manually provided by a human expert, ii) automatically derived from engineering data 116 (as further described below), or any combination of the two methods. The causal graph 118 describes how the variables PV in the process 102 may affect each other. In the causal graph 118, a directed edge from another variable B to variable A represents an expected causal relationship from B to A.

[0052] Using the causal graph 118, supplemented by the observation data 112, the decision support system 110 performs causal inference in three use cases to assist a human operator or an automated plant operating system: (1) finding root causes; (2) finding corrective actions; (3) what-if analysis to perform a sanity check before performing a corrective action. Causal inference is used to determine whether a causal hypothesis exists in the observation data.

[0053] Figure 2It is a flowchart showing a method 200 for performing causal inference. Performing causal inference includes: identifying 202 a causal effect 120 based on an input query 201; estimating 204 the causal effect; optionally verifying 206 the causal effect; and presenting 208 the causal effect.

[0054] In other words, in step 202, the causal graph 118 is used to determine which conditional effects (probabilistic influences) need to be considered when estimating the strength of the causal relationship in step 204. In step 206, the sensitivity of the estimate can be tested. At step 208, the method returns an indication as to whether variable A is actually affected by variable B.

[0055] Starting from the causal graph 118, describing the assumed causal relationships between process variables includes control variables such as setpoints or actuator values, and inferring that the causal relationships need to be examined. At this point, the input query 201 is provided in the form of "Does variable B affect variable A".

[0056] After finding all possible paths from B to A in the causal graph 118 in step 202, two conditional probability formulas are used in step 204 to determine whether there is a causal relationship between A and B:

[0057] P1 = P(A|B, confounders) and P2 = P(A|confounders), where the confounders are all process and controlled variables that affect A and B. If P1 is equal to P2, there is no causal relationship between A and B (each statistical relationship is explained by the confounders), and if P1 is not equal to P2, there is a causal relationship. The greater the difference between P1 and P2, the stronger the relationship. Possible methods for estimating the statistical significance of the difference between P1 and P2 are, for example, propensity-based stratification, propensity score matching, inverse propensity weighting, regression, regression discontinuity.

[0058] Step 206 tests the strength and validity of the causal relationship found in the previous step. Possible methods are to test the relationship using only a subset of the data (which should produce similar results) or (in an experiment) using a control treatment (replacing the data in which B is present with data in which B is absent).

[0059] For diagnosis and root cause analysis, the decision support system 110 uses the causal graph 118 and the observed data 112 to identify possible causes of anomalies in one or several process values. For each process value, one causal graph 118 can be used. Additionally or alternatively, there may be one causal graph covering the entire process 102.

[0060] To infer the possible root causes of anomalies or undesired behavior of certain process variables PV1…PV n according to the above regarding Figure 2Analyze the causal graph 118 of each process variable using the described method. The input query 201 is repeated for each variable B, where variable B is not equal to A = PV i 。By searching for possible causes on PV1…PV n Integrate the results of repeating the process on PV1…PV n to find the most likely root cause. Additionally, regression can be used to determine the direction of the effect.

[0061] In addition, topological data can be used to refine the root cause analysis. For example, in the case where root cause analysis identifies a variable in the causal graph (e.g., the controlled temperature) as the root cause, topological data can be used to identify factors at the asset level that affect the variable at the process level outside the causal graph (e.g., the motor of the cooling unit that controls the temperature).

[0062] To identify corrective actions, the causal graph 118 and the observed data 112 are used to identify the controlled process variables that exhibit a causal relationship with the measured process variables to be corrected.

[0063] To infer corrective actions, the process is similar to the above root cause analysis. The causal graph 118 of PV1…PV n (e.g., those process variables PV identified as possible root causes in the root cause analysis) to be manipulated is analyzed. At this point, instead of examining the causal relationships of all variables B!= PV i , only the controlled variables C (e.g., the setpoint of the control loop, the actuator value) that can be changed by the operator or the plant operating system are examined. Again, the results of PV1…PV n are integrated to find as few corrective actions as possible with the desired effect. Heuristic algorithms can be used to suggest or select operations. For example, setpoint changes (where the controller manages the desired effect) are more preferred compared to directly manipulating the actuator.

[0064] For what-if analysis (identifying side effects), the same causal graph 118 and the observed data 112 used to infer the root cause are used. In root cause analysis, we ask the question: Did the change observed in one variable cause the effect in another variable? However, in what-if analysis, the question we ask is: What if the value of a variable (i.e., the setpoint related to that variable) changes?

[0065] When variable B affects variable A, the degree of influence of B on A may vary for different values of a third variable C. Variable B can be a controlled variable that is controlled using a setpoint or actuator value, or a variable that is affected by such a controlled variable. In other words, the variation in the effect of variable B on variable A may vary depending on the different degrees of valve closure. To infer what happens to the effect of variable B on variable A when variable C is set to a specific value, the same causal diagram 118 and observational data 112 are used. The causal diagram 118 of process variable A is analyzed and P(A|do(B=b),C=c) is evaluated, where the value of "c" is fixed ("hypothetical value") and B is a variable that affects A but remains unchanged itself when C=c (i.e., B is not affected by C). For each causal diagram 118 in which there are variables, the input query 201 asks whether the controlled variable B has a causal relationship with the dependent variable A of the causal diagram 118 when C=c. The results for all diagrams constitute a list of possible side effects.

[0066] The steps can also be combined in a sequential manner:

[0067] (1) The (multiple) candidate root causes of the process anomaly are determined;

[0068] (2) The possible control variables that can resolve the root cause or symptom are identified (the process may favor corrective actions that have an expected effect on many possible root causes);

[0069] (3) If a how - analysis is performed, explain what may happen when the controlled variable is changed, and finally analyze the possible side effects of the action. Corrective actions with fewer side effects can be preferred. Additionally or alternatively, each dependent variable PV can be assigned a critical value such that corrective actions with little side effect on the critical process variable PV are preferred. The results of the individual steps can be presented to the operator or used as candidate actions for an automated plant operating system such as a plant operation agent based on reinforcement learning.

[0070] Now reference will be made to Figure 3 and Figure 4 to describe an example use case, Figure 3 showing a topology diagram of a data center, Figure 4 illustrating a causal diagram derived from the topology diagram or learned from engineering data (e.g., by using propensity - based techniques, covariate matching, etc.) or using both SME knowledge and engineering data with the help of a Subject Matter Expert (SME). The causal diagram is a directed acyclic graph, including a set of vertices and a set of directed edges between the vertices.

[0071] A data center has server racks and air conditioning units, which can push cold air to regulate the temperature of the servers. Sometimes, hotspots may occur in the aisles of the data center. There can be various reasons for hotspots. For example, a malfunction in the motor powering the cooling fans that regulate the cold air flow may result in a lack of cold air in the data center aisle, leading to the generation of hotspots. Or, the chiller may not be operating properly, and the supply air temperature may not be cold enough to avoid hotspots in the aisle. Similarly, if some servers in the aisle are overloaded due to unfair application deployment, hotspots may be created.

[0072] In the first example use case, Root Cause Analysis (RCA) is performed. A hotspot is detected in the aisle temperature measured near a set of servers. The RCA input query is formulated as: Does the cooling unit fan speed (variable B) affect the aisle temperature (variable A)? Since there are no confounding factors in this example, the variables that the root cause analysis may be interested in are:

[0073] · Supply air flow, cooling unit fan speed;

[0074] · Supply air temperature, chiller temperature;

[0075] · Server load, application deployment.

[0076] Given a topology graph and a causal graph, we analyze the causal graph of the affected variables (i.e., the causal graph of the aisle temperature) to perform further analysis on each causal variable. For convenience, assuming a constant server load, statistical estimation can be performed on the observed data of the following variables:

[0077] Supply air flow, Pr(hotspot|do(SFR=fr));

[0078] Supply air temperature, Pr(hotspot|do(SAT=t)).

[0079] The values “fr” and “t” can be selected based on operator experience and / or using values near known set points to study what would happen if the set points were changed. A list of the most likely root causes is obtained, ranked by probability, indicating, for example, that the cooling unit fan motor may not be working properly or may not be well regulated.

[0080] In the second example use case, a corrective action is sought. The input query here can take the following form: Does a change in the set value of the supply air flow (control variable) affect the aisle temperature (hotspot)? The probability analysis process is similar to RCA, except that only the control variable is considered in order to find a corrective action. Continue Figure 3 and Figure 4In the example shown in [reference], based on the causal diagram, the causal relationship can be estimated as in the RCA example, and then the supply air temperature setpoint can be lowered, or the supply air flow setpoint can be increased.

[0081] In the third example use case, if an analysis is performed to determine the extent to which the causal effect of the supply air flow on the hot spot is modified by the supply air temperature. Due to changes in the cooling unit fan speed, changes in the supply air flow (SFR) can be observed. The causal diagram is observed for the hot spot (i.e., the channel temperature). Then Pr(hotspot|do(SFR = f), SAT = t) is estimated for all possible supply air temperature (SAT) values. The value "f" can again correspond to a value at or near the setpoint.

[0082] Referring again to Figure 1 , a modeling tool 114 is also shown for constructing a causal diagram 118 using engineering data 116 and observational data 112. The causal diagram 118 is then provided to the decision support system 110.

[0083] The causal diagram 118 captures the relationships between the measured variables and the controlled variables in the process 102. There may be more than one causal diagram 118 to capture the process 102. For example, one diagram for each process variable PV (including process KPIs) models each PV as a dependent variable. The information provided in the causal diagram 118 can be qualitative (a relationship exists) rather than quantitative (how strong the relationship is and whether it is positive or negative). According to the present disclosure, a graph-based model is first used to select a relevant set of measurements (e.g., from hundreds of candidates to a few), and then the above method is performed to refine the network structure and estimate the strength of the relationship according to the conditional probability.

[0084] Now referring to Figure 5 , the modeling tool 114 can be configured to combine engineering data 116 (e.g., plant topology) and process model data with observational data 112 from process control and monitoring. More specifically, the modeling tool 114 can combine process topology information and process control configuration with information about process state (including past and present measurements), history, and alarm and event data. In this way, the dynamic information from the process control system 108 and the static information about the process topology (e.g., possible information, material, or energy flows in the process 102) can be combined. Preferably, the process topology information is available in the latest machine-readable form.

[0085] The modeling tool 114 can be configured (using, for example, a programming interface to relevant components of the process control system 108) to:

[0086] 1. Find the connections between elements (material, energy, and information flows) in the process topology;

[0087] 2. Use the following to find connections between control system objects, such as control modules, I / O objects / labels, aspect objects, alarm conditions, events, historical measurements and events / time series, and objects in a topology model:

[0088] a) Naming conventions, such as tag names;

[0089] b) Information available in process control application configurations, such as connections between process variables, I / O points, and characteristics in (e.g.) OPC DA;

[0090] c) Information available in process history configurations;

[0091] d) Specific conventions used in control application libraries can be used (a shared knowledge base of such conventions for the control application library available for Plant 100 can be established over time);

[0092] 3. Use the following to select a relevant subset of process variables / alarm conditions for a specific (abnormal) situation:

[0093] a) Process control system state information, such as: a set of process variables or alarm conditions, as a starting point;

[0094] b) Process topology model, as a graph, containing topology elements (instruments, control loops, pipes, vessels, etc.) and their connections (material, information, energy flows);

[0095] c) Use graph analysis algorithms and heuristics that traverse the topology model based on edges existing between plant elements nodes being material flows or information flows in a selected direction (i.e., forward or backward) or by default bidirectionally. The algorithms can use a reference model that defines the types of topology nodes and edges and the hierarchical relationships of node types;

[0096] 4. Add historical context, such as event history and past process variable measurements of a selected subset of process variables / alarm conditions, using observed data 112 for example (this information can be sorted according to its time or frequency, e.g., the most frequently occurring alarms during the process under investigation can be highlighted using a color scale or a numeric weight to indicate);

[0097] 5. Create a visualization of a specific situation for the operator of Plant 100 using the resulting data set, especially in the form of a causal graph, where the nodes in the updated conditional probability link graph represent, for example, plant elements (vessels, pipes, valves, control devices, instruments, etc.). Visual effects such as color, thickness, edge animation, etc. can be used to represent the likelihood value of one variable affecting another or to emphasize links related to the query variable (e.g., the variable showing abnormal behavior). Example queries are as described above.

[0098] The resulting causal graph can be used as an input for the above-described tool-based analysis of the plant 100, or provide data for time series analysis or machine learning algorithms for, e.g., root cause and fault path detection. For example, a topology-based causal graph can be used to perform computationally expensive analyses such as transfer entropy, or create predictive models that determine the impact in a given situation.

[0099] In this way, a (qualitative) model of the process 102 can be created without relying on the expertise of an SME, whose expertise can help predict the consequences of possible operator actions on relevant parts of the process 102.

[0100] The model (e.g., causal graph) can also be used to parameterize and initialize (quantitatively) a simulation model, which can provide impact calculations in the form of predicting the consequences of possible operator actions on relevant parts of the process 102. The simulation model can be used to identify possible user inputs that may change the current situation (e.g., plant elements that have a causal impact on an abnormal situation derived from the causal graph). A list of these input values can be used to create simulation experiments in which these values are changed to quantify the impact on the process state, e.g., to avoid triggering alarm limits and system trips.

[0101] The modeling tool 114 and / or the decision support system 100 can allow a user to manually edit the displayed results and provide feedback related to the automated analysis of abnormal situations and their root causes. Logs of user interactions with the system can be mined for performance improvement and better meeting of user expectations.

[0102] The causal graph created in this way can be used in the above-described manner to identify parts of the process state related to a particular situation under investigation. This subset of the available process state information augmented with relevant topology information can then be used to guide and focus the above-described tool-assisted situation analysis to efficiently and effectively solve the root cause problem. Parts of the process state may involve topology information and / or information such as system alarms or signal / measurement values from control loops or instrumentation related to the particular situation. "Related" can be defined as that part being associated with the situation (i.e., abnormal signal shape or alarm) with a probability higher than a corresponding threshold probability.

[0103] Although the decision support system 110 and the modeling tool 114 are shown as separate, distinct units, it should be understood that these systems can form part of the same unit such as a process control system, and these systems can be distributed systems.

[0104] Now refer to Figure 6, which illustrates a high-level diagram of an exemplary computing device 800 that can be used in accordance with the systems and methods disclosed herein. The computing device 800 includes at least one processor 802 that executes instructions stored in a memory 804. The instructions can be, for example, instructions for implementing the functions described as being performed by one or more of the above components or instructions for implementing one or more of the above methods. The processor 802 can access the memory 804 via a system bus 806. In addition to storing executable instructions, the memory 804 can also store session inputs, scores assigned to the session inputs, and the like.

[0105] The computing device 800 additionally includes a data repository 808 that can be accessed by the processor 802 via the system bus 806. The data repository 808 can include executable instructions, log data, and the like. The computing device 800 also includes an input interface 810 that allows external devices to communicate with the computing device 800. For example, the input interface 810 can be used to receive instructions from an external computer device, a user, and the like. The computing device 800 also includes an output interface 812 that interfaces the computing device 800 with one or more external devices. For example, the computing device 800 can display text, images, and the like via the output interface 812.

[0106] External devices contemplated to communicate with the computing device 800 via the input interface 810 and the output interface 812 can be included in an environment that provides substantially any type of user interface with which a user can interact. Examples of user interface types include graphical user interfaces, natural user interfaces, and the like. For example, a graphical user interface can accept input from a user using input devices such as a keyboard, a mouse, a remote control, etc., and provide output on an output device such as a display. Additionally, a natural user interface can enable a user to interact with the computing device 800 in a manner that is not constrained by input devices such as a keyboard, a mouse, a remote control, etc. Instead, a natural user interface can rely on speech recognition, touch and stylus recognition, gesture recognition on and near the screen, air gestures, head and eye tracking, sound and voice, vision, touch, pose, machine intelligence, and the like.

[0107] Additionally, although shown as a single system, it should be understood that the computing device 800 can be a distributed system. Thus, for example, several devices can communicate via a network connection and can jointly perform the tasks described as being performed by the computing device 800.

[0108] The various functions described herein can be implemented using hardware, software, or any combination thereof. If implemented in software, these functions can be stored or transmitted as one or more instructions or code on a computer-readable medium. A computer-readable medium includes a computer-readable storage medium. A computer-readable storage medium can be any available storage medium accessible by a computer. By way of example and not limitation, such a computer-readable storage medium can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store instructions or data structures in the form of and accessible by a computer. As used herein, disk and optical disks include compact disk (CD), laser disk, optical disk, digital versatile disk (DVD), floppy disk, and Blu-ray disk (BD), where disks typically reproduce data magnetically, while optical disks typically reproduce data optically using lasers. Additionally, propagated signals are not included within the scope of computer-readable storage media. A computer-readable medium also includes a communication medium, which includes any medium that facilitates the transfer of a computer program from one place to another. For example, a connection can be a communication medium. For example, if software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are all included in the definition of a communication medium. Combinations of the above should also be included within the scope of computer-readable media.

[0109] Alternatively or additionally, the functions described herein can be performed, at least in part, by one or more hardware logic components. By way of example, and not limitation, the types of illustrative hardware logic components that can be used include field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), system on a chip (SOC), complex programmable logic devices (CPLD), and the like.

[0110] [Template]

[0111] The applicant hereby separately discloses each individual feature described herein and any combination of two or more such features, provided that regardless of whether such features or combinations of features solve any of the problems disclosed herein, and without limiting the scope of the claims, such features or combinations can be implemented in their entirety based on the general knowledge of those skilled in the art from this specification. The applicant points out that aspects of the present invention can consist of any such individual feature or combination of features.

[0112] It should be noted that the embodiments of the present invention are described with reference to different categories. Specifically, some examples are described with reference to methods, while other examples are described with reference to devices. However, those skilled in the art will learn from the description that, unless otherwise stated, any combination between features related to different categories is also considered to be disclosed in the present application, in addition to any combination of features belonging to one category. However, all features can be combined to provide a synergistic effect, rather than just a simple sum of the features.

[0113] Although the present invention has been illustrated and described in detail in the drawings and the foregoing description, such illustration and description are to be considered illustrative rather than restrictive. The present invention is not limited to the disclosed embodiments. By studying the drawings, the disclosure, and the appended claims, those skilled in the art can understand and implement other variations of the disclosed embodiments.

[0114] The word "comprising" does not exclude other elements or steps.

[0115] The indefinite article "a" or "an" does not exclude a plurality. Additionally, unless otherwise stated or clearly indicated as singular from the context, the articles "a" and "an" as used herein shall generally be construed to mean "one or more".

[0116] A single processor or other unit can implement the functions of several items recited in the claims.

[0117] The fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used advantageously.

[0118] Any reference signs in the claims shall not be construed as limiting the scope.

[0119] Unless otherwise stated or clear from the context, the phrase "A and / or B" as used herein is intended to represent all possible permutations of one or more of the listed items. That is, the phrase "X includes A and / or B" is satisfied by any of the following: X includes A; X includes B; or X includes both A and B.

Claims

1. A decision support system for an industrial plant, the decision support system being configured to: Obtain a causal graph that models a causal hypothesis related to conditional dependencies among variables in the industrial plant; Obtain observation data related to the operation of the industrial building; And Use the causal graph and the observational data to perform causal inference to estimate at least one causal effect related to making decisions when operating the industrial plant, Wherein performing the causal inference includes: identifying the causal effect based on an input query; estimating the causal effect; validating the causal effect; and presenting the causal effect, Wherein estimating the causal effect includes: using a conditional probability formula to estimate the strength of the causal relationship between two of the variables, Wherein validating the causal effect includes performing data subset validation by re - estimating the strength of the causal relationship using only a subset of the observational data and / or performing a control treatment by re - estimating the strength of the causal relationship using the observational data, wherein data including one variable is replaced with data where the same variable does not exist.

2. The decision support system according to claim 1, wherein presenting the causal effect includes: Output the causal effect to a human plant operator and / or output the causal effect to an automated plant operating system.

3. The decision support system according to any one of the preceding claims is further configured to perform a root cause analysis related to a given variable in the industrial plant to identify which of the other variables in the causal graph are most likely to affect the given variable, wherein performing the root cause analysis on the given variable includes: Perform the causal inference to estimate the strength of the causal relationship between the given variable and each of the other variables.

4. The decision support system according to claim 1 or 2, further configured to find corrective actions that can affect a change in a given variable, wherein finding the corrective actions includes: Use the causal graph and the observational data to identify controlled variables in the industrial plant, the controlled variables exhibiting a causal relationship with the given variable, and perform the causal inference to estimate the strength of the causal relationship between the given variable and each of the other variables that are controlled variables in the causal graph.

5. The decision support system according to claim 1 or 2, further configured to perform a what - if analysis to identify one or more possible side - effects of changing a controlled variable in the industrial plant.

6. The decision support system according to claim 5, wherein performing the what-if analysis includes: Identify a first variable and a second variable in the causal graph, the first variable and the second variable each exhibiting a causal relationship with a third variable rather than a causal relationship with each other; And perform the causal inference to estimate the degree to which the causal effect of the first variable on the third variable is modified by the second variable.

7. A decision support method for an industrial plant, the decision support method including: Obtain a causal graph that models a causal hypothesis related to conditional dependencies among variables in the industrial plant; Receive observational data related to the operation of the industrial plant; And Use the causal graph and the observational data to perform causal inference to estimate at least one causal effect related to making decisions when operating the industrial plant, Characterized in that Performing the causal inference includes: identifying the causal effect based on an input query; estimating the causal effect; validating the causal effect; and presenting the causal effect, Wherein estimating the causal effect includes: using a conditional probability formula to estimate the strength of the causal relationship between two of the variables, Wherein validating the causal effect includes performing data subset validation by re-estimating the strength of the causal relationship using only a subset of the observational data and / or performing a control process by re-estimating the strength of the causal relationship using the observational data, wherein data including one variable is replaced with data in which the same variable does not exist.

8. A computer-readable medium comprising instructions that, when executed by a computing device, cause the computing device to perform the method according to claim 7.

Images