Monitoring is a critical task to manage modern applications using cyber-physical systems such as in healthcare, environmental monitoring, water and energy distribution or Industry 4.0. Its main goal is the optimization of the equipment use, the prediction or avoidance of failures, and the change in the operating modes of the system according to both the system goals and its current condition. The abnormal situations that could lead to failures can have different levels of severity, and can be nested in different ways. In this context, this paper proposes a method to build a lattice, ordering those situations depending on the constraints they rely on. This lattice represents a road-map of all the situations that can be reached from a given one, desirable or undesirable. This helps in decision support, by allowing the identification of the actions that can be taken to correct the abnormality avoiding in this way the interruption of the system processes.
Monitoring is a critical task to manage cyber-physical systems in different application fields such as health and environmental monitoring, or water and energy distribution. Its goal is to analyze an ongoing process to find out whether it behaves according to expectations [12]. This allows the avoidance or prediction of undesirable behaviors or situations, such as failures, and the change in the operating modes of the system according to both the system goals and its current condition. Such is also the case of Industry 4.0, whose main objective is to improve production and associated services through the digitization and automation of manufacturing processes. Several fields and technologies, such as the Internet of Things, Robotics, Cyber-Physical Systems, are fundamental to Industry 4.0 in order to build intelligent machines, storage systems and production facilities capable of exchanging information in an autonomous and smart way.
Early detection of situations that may lead to failures in the Industry 4.0 scenario requires the integration of data from heterogeneous data sources in real-time. There are several works that propose the use of different technologies to deal with those issues.
Copyright c 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
The approach presented in [13] exploits big data technologies for data-driven anomaly
detection in manufacturing processes. In [
Manufacturing processes are not always executed in optimal conditions, without being stopped completely. Expert knowledge enables to describe these ”intermediate” manufacturing conditions. The associated abnormal situations can have different levels of severity, and be nested in different ways: they can impact other processes or resources that participate in these processes, they can trigger other situations that represent a risk of major interruption of the manufacturing process or a risk of accident.
Once a situation that may lead to failures is detected, decisions must be taken, such as deciding whether the process should be interrupted or continued under sub-optimal conditions. In order to help in decision making and choose the most appropriate action, it is relevant to consider which other situations can be reached according to the possible actions. Thus, this paper proposes an approach to establish an order among the situations, depending on how their constraints are correlated. This order represents a road-map of all the situations, desirable or undesirable, that can be reached from a given one. In this way, it is possible to identify the actions that can be taken to correct the abnormality, considering that certain actions can modify the value of a property and thus change the state of the system, either by satisfying another constraint or, on the contrary, by not satisfying constraints anymore.
The proposed approach uses the lattice theory [
The remainder of the paper is structured as follows: in section 2 the related work is presented. In section 3, the general approach is introduced, providing the definitions for the construction of the hierarchy of situations. In section 4, the interpretation and exploitation of the lattice to support the decision making is detailed. In Section 5, we present some concluding remarks, including some perspectives for future work. 2
As mentioned in the introduction, the definition of abnormal situations requires the integration of data from different data sources, with different underlying meanings and different temporal resolutions. Furthermore, these situations may share constraints and involve similar resources. These challenges have been addressed from different areas of research.
Most of the proposals coming from the Complex Event Processing (CEP)
community [
Other approaches to structurally define the event type composition in cyber-physical systems (CPS) adopt the theory of concept lattice [16]. Concept lattice has been used in machine learning, knowledge discovery and software engineering. In [14], a concept lattice-based event model for CPS is presented. With this model, a CPS event is uniformly represented by three components: the event type, its internal attributes, and its external attributes. The internal and external attributes together characterize the event type. The model allows events to be composed across different components and devices within and among both the cybernetic and physical domains.
Another approach is the use of rule-based models such as Adaptive Neuro Fuzzy Interference Systems (ANFIS) models for monitoring wind turbine SCADA (Supervisory Control and Data Acquisition) signals [10,11]. In order to obtain turbine condition statements, the authors implement rules given by an expert who is familiar with the behavior of the turbine, typical faults and their root causes. There are two types of rules: generic rules used to highlight anomalies, and specific rules providing specific condition or potential root cause. In this case, it is possible to determine whether one antecedent is contained in another or not. However, there is not a direct way to determine that an antecedent or situation is partially occurring or that other violations must be satisfied for another situation to occur.
In most of the solutions mentioned above, it is possible to observe a hierarchy of events in the sense that complex events are composed of simple ones, but this hierarchy does not directly provide the information that certain complex events may occur partially in other complex events or that they share certain simple events or not. To the best of our knowledge, no works have been done that attempt to establish a formal representation of a hierarchy among (abnormal) situations. For this, establishing an order among situations permits taking into account the relations that exist among each other and to support the actions to take to overcome the abnormal situations when they happen.
The contribution of the approach presented in this paper is a method that allows to order not only the complex events already defined based on the simple events that compose them, but also to order certain combinations of simple events that do not completely compose a complex event. Additionally, it is possible to consider occurrences of simple events that imply the occurrence of other simple events for the construction of the hierarchy of situations. 3
The idea driving this approach is to formally represent a hierarchy among situations that may lead to failures. Therefore, through the use of this hierarchy of situations, decisions can be made based on which others situation can be reached according to the constraints that can be satisfied. This allows to adapt the maintenance schedule or to take further measures to prevent unexpected downtime.
R c.1
In this section, we firstly introduce the definitions that are necessary for the construction of the hierarchy of situations, and secondly, we describe the steps for building the hierarchy. In order to formally represent a hierarchy of situations, let us consider the following structure hS ; C; R; T i where: – S = fs1; s2; s3; : : : ; sng is the set of all the situations, – C = fc1; c2; c3; : : : ; cmg is the set of all the constraints, – R S C is a binary relation that links a situation with a constraint and – T C C is a binary relation that links a constraint with another constraint.
A situation defines an abstract state of affairs that represents a particular scenario of interest and involves observations linked through spatio-temporal relationships, resources and processes. Let us note that situations are abstract, meaning that there may be different instances of a given situation. Instances of the same situation can happen in different periods of time, but they all satisfy the same constraints.
The set C contains all the constraints that are associated with one or more situations in the set of situations S . These constraints concern different properties of the processes, machines, resources or even the environment in which the tasks are executed. For example, if we consider the variables MC1 Temp and MC2 Temp, corresponding to the temperature of a component of a machine and the temperature of another component of the same machine, then MC1 Temp < 40 C and MC1 Temp>MC2 Temp are constraints defined on them.
The binary relation R is used to establish that a constraint is involved in a situation. We write s1Rc1 to indicate that the constraint c1 is involved in the situation s1. The R relation is therefore built from the relationships between the situations and the constraints that are extracted from expert knowledge.
For example, consider the following set of constraints and set of situations, with six constraints and six situations:
C = fc1, c2, c3, c4, c5, c6g and S = fs1, s2, s3, s4, s5, s6g.
Constraints The corresponding R relation, built from the set of situations S and the set of constraints C, is shown in Table 1 (columns 1 and 2). For example, the 4th line in this table indicates that if constraints c2 and c5 are satisfied (2nd column) then situation s4 happens (1st column).
Another way to show relation R is depicted in Figure 1. In this figure, we can observe how different situations share a part of the constraints in their definition, for example s1 and s2 share constraints c1 and c6.
Some constraints can be more general than others, i.e. include others. Such is the case with, for example, the constraints c1 and c2 defined as MC1 Temp < 40 C and MC1 Temp < 60 C, respectively. If c1 is satisfied, then c2 is necessarily also satisfied. Furthermore, some constraints may imply other constraints due to physical properties extracted from expert knowledge or observations. For this reason, the T relation is defined to indicate that if a constraint is satisfied, then another one is also satisfied. We write c1T c2.
For example, let us consider the implications among the constraints from the set of constraints C, presented before, shown in Figure 1 (arrows labeled with T ). These implications are inferred either from mathematical properties, or from observations, or extracted from expert knowledge. Taking into account these relations, the constraints associated with each situation are established as shown in Table 1 (third column). This allows to associate more constraints than those that are explicitly concerned by the situations.
In order to formalize the relations between situations and constraints, two operators are defined below, based on the use of both R and T relations. These operators are defined on a set of situations or constraints because each situation can involve several constraints, and several situations can have several constraints in common.
The first operator dX e enables the retrieval of the set of constraints associated to a set of situations.
dX e := fc 2 Cj8x 2 X : xRc _ 9c0 2 C : xRc0 ^ c0T cg
Considering the situations s1 and s2 of the example presented above, the constraints involved in both situations are dfs1; s2ge = fc1; c2; c4; c5; c6g.
The second operator bYc conversely enables the retrieval of the set of situations involving a set of constraints Y.
bYc := fs 2 Sj8y 2 Y : sRy _ 9c0 2 C : sRc0 ^ c0T cg
This operator retrieves the situations involving at least all the constraints in the set Y. Therefore, considering the example presented above, if the constraints c2 and c5 are chosen then the situations involving those constraints are bfc2; c5gc = fs1; s2; s3; s4g. Let us note that these situations may involve other constraints, e.g. s3 with c4 and c6.
Using the elements of the structure hS; C; R; T i and the two operators d:e and b:c previously defined, the construction of the lattice representing the hierarchy of situations is detailed below. 3.2
First the nodes of the lattice are described. As highlighted above, situations have common constraints, nested in different ways: inclusion, non-null intersections, etc. Therefore, we propose to group them considering the constraints they share through the operators defined in the previous section.
Algorithm 1 Calculate all the pairs (X ; Y) where X bYc = X (setof P airs)
In our approach, we group situations and their constraints as pairs (X ; Y ) where X is a set of situations and Y is a set of constraints such that dX e = Y and bY c = X . The first component of the pair is a set with all the situations that share all the constraints of the second component. The second component is the set with all the constraints in common among the situations from the first component.
In order to find all the pairs and thus the nodes of the lattice, given a set of situations S, a set of constraints C, and relations R and T , Algorithm 1 is followed. Firstly, dfsge is computed for each situation s 2 S (lines 3-5). Then, for any two sets in this set of sets (consSet), their intersection is calculated. If this intersection is not yet contained in the set, it is added (lines 6-12). This step is repeated until no new sets are generated. If the set C of all the constraints and the empty set (fg) are not in consSet, they are added (lines 13-18). These sets yield the minimum and maximum nodes of the lattice, respectively. Finally, for every set O in consSet, bOc is computed (lines 19-21). By the end, the set (setof P airs) of all the possible (X ; Y ) pairs that satisfy dX e = Y and bY c = X are obtained.
Considering the example in Figure 1 and Table 1, if the set of situations S and the set of constraints C are given as input to the algorithm, then the output are the nodes of the lattice shown in Figure 2. Let us note that since the situations are predefined, the search for all pairs is done offline. If a situation is added, it is either added to a node or a new node is created.
Once all the pairs are found, the next step is to order those pairs in a lattice to build the hierarchy of situations. A lattice is an algebraic structure that consists of a partially ordered set in which every two elements have a unique supremum (also called a least upper bound or join) and a unique infimum (also called a greatest lower bound or meet). A partial order is a pair (P ; ) where P is a set and that it is reflexive, anti-symmetric and transitive.
For our lattice, we consider the set:
L = f(X ; Y )jX S ^ Y C ^ dX e = Y ^ bY c = X g and the following binary relation defined over it: is binary relation over P so
3.3
Proof. The proof is done in two parts. In the first part it is proved that (L; ) is a partially ordered set. In the second part, it is proved that for any two elements of the lattice they have a unique supremum and a unique infimum.
Part 1. First it is proven that is reflexive, i.e. 8(X ; Y ) 2 L j (X ; Y ) (X ; Y ). By the definition of , (X ; Y ) (X ; Y ) , X X ^ Y Y It is true since each set is self-contained. Then, is reflexive.
Now we prove that is anti-symmetric, i.e. 8(X ; Y ); (X 0; Y 0) 2 Lj(X ; Y ) (X 0; Y 0) ^ (X 0; Y 0) (X ; Y ) ) (X ; Y ) = (X 0; Y 0).
Let two sets (X ; Y ), (X 0; Y 0) belong to L, and (X ; Y ) (X 0; Y 0) and (X 0; Y 0) (X ; Y ). Applying the definition of on both we obtain: X X 0 ^ Y 0 Y and X 0 X ^ Y Y 0, respectively. Thus, on the one hand X X 0 and X 0 X means that X = X 0, and on the other hand Y 0 Y and Y Y 0 means that Y = Y 0. Since (X ; Y ) = (X 0; Y 0), is anti symmetric.
Finally, we show that is transitive, i.e. (X ; Y ) (X 0; Y 0) ^ (X 0; Y 0) (X 00; Y 00) ) (X ; Y ) (X 00; Y 00).
Let (X ; Y ), (X 0; Y 0); (X 00; Y 00) belong to L, and consider that (X ; Y ) (X 0; Y 0) and (X 0; Y 0) (X 00; Y 00). Now, if we apply the definition of on both we get, X X 0 ^ Y 0 Y and X 0 X 00 ^ Y 00 Y 0, respectively. Considering sets X , X 0, and X 00 X X 0 and X 0 X 00 means that X X 00 because is transitive. Similarly, Y 00 Y 0 and Y 0 Y means that Y 00 Y . Then, (X ; Y ) (X 00; Y 00). Therefore, is transitive.
Part 2. Firstly, two lemmas are introduced:
Y dbY ce and X bdX ec.
Proof. bY c has all the situations involving the constraints in the set Y and the situations involving constraints that are included by the constraints in the set Y . Then, dbY ce has all the constraints associated with the set of situations bY c. Therefore, the set of constraints Y is contained in dbY ce. The reasoning is the same for the set of situationsX .
Y
Y0 ) bY0c bYc and X
X 0 ) dX 0e dX e.
Proof. We prove that dX 0e dX e. By definition of d:e on X 0, dX 0e = fc 2 Cj8x 2 X 0 : xRc _ 9c0 2 C : xRc0 ^ c0T cg This is for all b 2 X 0, and by hypothesis X X 0. Therefore, dX 0e = fc 2 Cj8x 2 X 0 : xRc _ 9c0 2 C : xRc0 ^ c0T cg
fc 2 Cj8x 2 X : xRc _ 9c0 2 C : xRc0 ^ c0T cg = dX e The same reasoning can be applied to prove Y Y0 ) bY0c bYc. This lemma shows that the more constraints are required, the fewer situations involve all of them. Conversely, the more situations we consider, the fewer constraints they have in common.
Having defined the two lemmas and their respective proofs, we continue with the general proof of the theorem.
For any two pairs (X ; Y) and (X 0; Y0) we obtain: – the infimum (greatest common pair) of (X ; Y) and (X 0; Y0) as
(X ; Y) ^ (X 0; Y0) := (X \ X 0; dbY [ Y0ce) – the supremum (least common pair) of (X ; Y) and (X 0; Y0) as
(X ; Y) _ (X 0; Y0) := (bdX [ X 0ec; Y \ Y0)
To prove the existence of the unique infimum, we have to show that (X ; Y) ^ (X 0; Y0) is smaller than both (X ; Y) and (X 0; Y0), and any other common child of (X ; Y) and (X 0; Y0) is also a child of (X ; Y) ^ (X 0; Y0).
First we prove that (X \ X 0; dbY [ Y0ce) (X ; Y). By the definition of , we have X \ X 0 X and Y dbY [ Y 0ce.
The proof of X \ X 0 X is trivial. For the other part of the conjunction we start from Y Y [ Y0 applying Lemma 2 twice we obtain dbYce dbY [ Y 0ce. In addition, Lemma 1 expresses that Y dbYce. Thus, combining the last two statements we conclude that Y dbYce dbY [ Y 0ce. Therefore, Y dbY [ Y 0ce.
The same method can be used to prove that (X \ X 0; dbY [ Y0ce) (X 0; Y0).
Similarly, (X ; Y) _ (X 0; Y0) is greater than both (X ; Y) and (X 0; Y0), and it is a child of any common parent of these two pairs. Only the proof for the infimum is developed here since the proof for the supremum is similar. 4
An illustrative case study is described below and is used to highlight how the lattice can be used and the advantages it offers to support the decisions that need to be made when an abnormal situation is detected in an industrial framework.
The case study is based on a manufacturing production line composed of several machines. These machines are equipped with sensors on different components. The
Set of constraints C
sensors collect data on the properties described in Table 2. This scenario is formally
represented using the manufacturing domain ontology introduced in [
Several abnormal situations that could lead to failures in this scenario are defined from expert knowledge. Each of them is expressed as a set of constraints. For reasons of space, situations are not described in detail, but the situations represented here cover the following types of failures: hydraulic oil leakage, cooling system filter obstructions, converter and rotor malfunctions and global malfunctions of the production line. The defined abnormal situations are shown in Table 3 with a brief description of what they represent; and the constraints concerned by them are described in Table 2.
Some situations represent conditions that can lead to the same potential failure, indicating different levels of severity. For example, situations s1 and s2 both represent situations that could lead to the same failure but s2 indicates a higher severity since the temperature threshold is higher than the temperature threshold for s1. In this case, the actions to be taken may be more decisive or follow a safety protocol, since the situations are of higher severity.
The resulting lattice from the scenario described above is shown in Figure 3. By observing the hierarchy of situations, it is possible to notice that in the upper part of the lattice the constraints which are involved in most of the situations are verified. The further we go down in the lattice, the more the situations are specific as they include more constraints. The stronger constraints are close to the bottom of the diagram, meaning that these situations embed those situated higher up in the lattice.
The lattice provides a structure that represents the order in which the situations may arise according to which constraints are verified. Considering that certain actions can modify the value of a property and thus change the state of the system, either by satisfying another constraint or on the contrary by not satisfying a constraint anymore, the lattice allows the analysis of the actions to take. It implies, from the decision support point of view, reaching a node situated lower in the lattice if new constraints are satisfied, or higher in the other case.
For example, if the constraints c15, c17 and c18 are satisfied then it means that the situation s6 is happening, i.e. there is a cooling filter obstruction in the machine M3. The lattice, and in particular the node (fs6; s7; s10g; fc15; c17; c18g); shows that situation s7, a more critical situation than s6, is reachable if no action is taken or if the actions taken imply the satisfaction of c16. In general, the actions aim at correcting the abnormal property values causing them to return to normal values. In this case, a filter change action in the cooling system would make the Converter water temp. and Generator temp. properties decrease to values lower than their respective thresholds, i.e. the constraints c15 and c18 would no longer be satisfied. This would lead to the process at the node (fs6; s7; s8; s9; s10g; fc17g) where only the constraint c17 is satisfied and other situations can be reached from there. Ideally, it is always intended to go up on the lattice to the node (S ; fg) where none of the constraints are satisfied, meaning that no abnormal situation is (partially) present.
A particular node in the lattice represents a possible state of the system. It should be noted that the minimum of the lattice, the node (fg; C), may represent an unreachable state. That is, since this node includes all the constraints (C), it may be that two constraints cannot be satisfied at the same time because they are exclusive, such as the constraints c6 and c7. However, this node is necessary for the hierarchy to be a lattice. If this happens in another node of the lattice that it is not the minimum, then there is a problem in the definition of the R relation or of the T relation: it would be a situation that can never be satisfied because it concerns constraints that are mutually exclusive. ({s1,s3,s2,s5,s4,s7,s6,s9,s8,s12,s11,s14,s13,s10},{ })
({s7,s6,s9,s8,s10},{c17}) ({s3,s5,s4},{c11,c8,c10}) ({s9,s8,s10},{c19,c17}) ({s12,s11,s14,s13},{c23,c21}) ({s1,s3,s2,s4,s9,s8},{c6}) ({s7,s6,s10},{c15,c17,c18}) ({s5,s4},{c9,c11,c8,c10}) ({s1,s2},{c1,c3,c5,c4,c6})
({s7,s10},{c16,c15,c17,c18}) ({s14,s13},{c13,c23,c21}) ({s3,s4},{c6,c11,c8,c10}) ({s9,s8},{c19,c17,c6}) ({s12,s11},{c12,c23,c21}) ({s5},{c7,c9,c11,c8,c10})
({s2},{c1,c2,c3,c5,c4,c6}) ({s10},{c18,c15,c16,c17,c19}) ({s14},{c14,c13,c23,c21}) ({s4},{c9,c6,c11,c8,c10}) ({s9},{c20,c19,c17,c6}) ({s12},{c22,c12,c23,c21})
For each node defined as (X; Y), all situations in X are at least partially occurring, i.e. a part of their constraints is satisfied. When one of the situations in X involves exactly all the constraints in Y and no other, this means that this situation is occurring. This is formally defined in Definition 4 (these situations appear in bold face in Figures 2 and 3).
k(X; Y)k := fs 2 Xj8y 2 Y : sRy ^ @c0 2 C-Y : sRc0g
For example, k(fs6; s7; s10g; fc15; c17; c18g)k = fs6g means that the situation s6 happens in the node (fs6; s7; s10g; fc15; c17; c18g) because it involves only the constraints c15, c17 and c18.
Regarding the nodes where k(X; Y)k is empty, this means that situations in X are partially occurring or that they are potential situations that the system could reach. The discovery of these combinations of constraints in common among certain situations can give rise to the definition of new situations that allow the early stage detection of certain ”relevant” situations. This would allow preventive decisions to be taken.
This paper presents an approach that uses the lattice theory to represent a hierarchical order among certain situations that lead to potential failures. This lattice is automatically built from existing knowledge and represents a road-map of all the situations that can be reached from a given one, desirable or not. This allows the identification of the actions that can be taken to correct the abnormality.
Although the case study presented in the paper comes from the Industry 4.0 context, the proposed approach can be applied to other application domains, where real time monitoring is needed. In fact, in any monitoring application, the notion of situation (normal or abnormal) exists. Their identification thanks to expert knowledge enables to automatically build the hierarchy among the identified situations using our approach.
It is also to be remarked that in the case study presented, the T relation represents a hierarchy between the constraints in the sense that one constraint may be more specific than another, so that when it is satisfied, the more general constraint is also satisfied. The T relation can also represent another type of relationship among the constraints such as the fact that two different properties have a physical link, meaning that the values of one directly affect the values of the other.
The following research lines will be addressed in future works. Firstly, a more extensive case with more complex situations, involving more constraints, will be explored. This raises scalability and complexity issues to build the lattice. Therefore, tests will be performed with different variants of Algorithm 1 to evaluate their relative efficiency. Secondly, a more exhaustive and detailed study will be made on the constraints implications that represent dependencies among constraints for the identification of new situations or the refinement of existing ones.
These PhD works are funded by the Normandy Region (France) in the framework of the STEaMINg (Semantic Time Evolving Models for Industry 4.0) project. 7. Gra¨tzer, G.: Lattice Theory: Foundation (01 2011). https://doi.org/10.1007/978-3-03480018-1 8. Nation, J.B.: Notes on lattice theory (1998) 9. Reis, M.S., Gins, G.: Industrial process monitoring in the big data/industry 4.0 era: From detection, to diagnosis, to prognosis. Processes 5(3), 35 (2017) 10. Schlechtingen, M., Santos, I.F.: Wind turbine condition monitoring based on scada data using normal behavior models. part 2: Application examples. Applied Soft Computing 14, 447 – 460 (2014). https://doi.org/https://doi.org/10.1016/j.asoc.2013.09.016, http://www. sciencedirect.com/science/article/pii/S1568494613003104 11. Schlechtingen, M., Santos, I.F., Achiche, S.: Wind turbine condition monitoring based on scada data using normal behavior models. part 1: System description. Applied Soft Computing 13(1), 259 – 270 (2013). https://doi.org/https://doi.org/10.1016/j.asoc.2012.08.033, http://www.sciencedirect.com/science/article/pii/S1568494612003821 12. Schreiber, G., Akkermans, H., Anjewierden, A., Hoog, R., Shadbolt, N., Velde, W., Wielinga, B.: Knowledge Engineering and Management - The CommonKADS Methodology, vol. 24 (01 2001). https://doi.org/10.7551/mitpress/4073.001.0001 13. Stojanovic, L., Dinic, M., Stojanovic, N., Stojadinovic, A.: Big-data-driven anomaly detection in industry (4.0): An approach and a case study. pp. 1647–1652 (12 2016). https://doi.org/10.1109/BigData.2016.7840777 14. Tan, Y., Vuran, M.C., Goddard, S., Yu, Y., Song, M., Ren, S.: A concept lattice-based event model for cyber-physical systems. In: Proceedings of the 1st ACM/IEEE International Conference on Cyber-Physical Systems. p. 50–60. ICCPS ’10, Association for Computing Machinery, New York, NY, USA (2010). https://doi.org/10.1145/1795194.1795202, https://doi.org/10.1145/1795194.1795202 15. Tommasini, R., Bonte, P., Della Valle, E., Mannens, E., Turck, F., Ongenae, F.: Towards ontology-based event processing. pp. 115–127 (02 2017). https://doi.org/10.1007/978-3-31954627-8 9 16. Wille, R.: Formal Concept Analysis as Mathematical Theory of Concepts and Concept Hierarchies, pp. 1–33. Springer Berlin Heidelberg, Berlin, Heidelberg (2005). https://doi.org/10.1007/11528784 1, https://doi.org/10.1007/11528784 1 17. Wu, E., Diao, Y., Rizvi, S.: High-performance complex event processing over streams.
In: Proceedings of the 2006 ACM SIGMOD International Conference on Management of Data. p. 407–418. SIGMOD ’06, Association for Computing Machinery, New York, NY, USA (2006). https://doi.org/10.1145/1142473.1142520, https://doi.org/10.1145/1142473. 1142520