5.1. Construction of Component Fault Mode Hypernetwork
The concept of the hypergraph was proposed by C. Berge in 1970. Hypergraphs can be used to analyze multi-layer heterogeneous networks. A network that can be described by a hypergraph is a hypernetwork [
27,
28]. A component of the machining center has multiple fault modes, and a fault mode may occur in multiple components. The components and fault modes are associated in the form of a set. The component fault mode network is a heterogeneous network composed of two types of nodes with different properties. Therefore, the machining center component fault mode hypernetwork model can be established to describe the interactions and effects between component fault modes.
During the operation of the machining center, different components may have different fault modes, and the same mode may occur multiple times in the same component. Therefore, this paper uses the number of modes occurring in the components as weights to build a weighted component fault mode hypergraph with components as hyperedges and fault modes as nodes.
Defining the component fault mode hypergraph as , then
- 1.
, is the number of fault modes and is the fault mode with code , called the node of the hypergraph.
- 2.
is a subset of , represents the component of the code , which is called hyperedge, and has the following properties:
- (1)
, which means that the component can have one or more modes.
- (2)
, the set of elements of all hyperedges in the hypergraph is the set of modes.
- (3)
means that fault mode is a unit of , indicating that the component has the th fault mode.
- (4)
If , node and are adjacent in component , denoted as . If there is no component making node adjacent to , denoted as .
- (5)
If , the hyperedge is adjacent to , denoted as , and the adjacent hyperedges have a common node. For , make , there is no isolated fault mode that does not belong to any component.
- 3.
represents the set of times the fault mode occurs in the component. is the weight of the node belonging to the hyperedge , if , the value of is not 0, otherwise it is 0.
- 4.
The fault mode node is described by the fault mode code and the fault mode occurrence probability . The larger is, the greater the probability of the fault mode occurring in the component is.
- 5.
Component hyperedge , described by code and reliability . The larger is, the higher the reliability of component is, and the smaller the probability of fault is.
- 6.
The component fault mode hypergraph can be expressed in two forms: set or matrix. Since the matrix can express the relationship between elements more clearly, this paper uses the form of matrix to describe the relationship between components and fault modes. The correlation matrix corresponding to the hypergraph is expressed by .
The row in the matrix corresponds to the component hyperedge , and the column corresponds to the fault mode node . Let denote the element in the th row and th column of the matrix.
The edge load attribute will be an important structural feature of key fault propagation path identification in subsequent systems. From the process of building a component fault mode hypernetwork model, combined with the collected fault information, a machining center weighted component fault mode hypergraph is obtained, as shown in
Figure 5.
According to
Figure 5, the correlation matrix
can be obtained as follows:
5.2. Propagation and Diffusion Influence Degree of Fault Mode
When one fault mode occurs in a component, it will cause the occurrence of other fault modes of the component, and it will diffuse to other components via the hyperedge connection in the hypernetwork, causing faults. The influence of a certain fault mode on the entire network has always been a concern of researchers. The calculation of the fault influence degree can quantitatively guide the risk evaluation. Therefore, on the basis of establishing the component fault mode hypernetwork, the fault propagation and diffusion influence degree are studied. It should be noted that the following definitions should be made when calculating the fault propagation and diffusion influence degree of components:
- 1.
The propagation influence of fault mode. If fault mode occurs and there is , the occurrence of fault mode will cause the occurrence of , and this process is called the propagation influence of fault mode.
- 2.
Fault propagation capability. The probability of occurrence of a fault mode in a component that causes other fault modes in the component to occur. The fault propagation capability is related to the occurrence probability of the mode .
- 3.
The influence degree of fault propagation. The propagation influence degree represents the probability that the occurrence of a certain fault mode will cause the occurrence of a specific fault mode. It is related to the propagation capability of the faults.
- 4.
The diffusion influence of fault mode. If , , and , the phenomenon that the occurrence of fault mode eventually leads to the occurrence of mode is called fault mode diffusion influence.
- 5.
The influence degree of fault diffusion. The diffusion influence degree indicates the probability that the occurrence of a certain fault mode will cause the occurrence of other fault modes.
- 6.
Fault diffusion trajectory. The diffusion trajectory from to is composed of a set of fault mode node sequences . Adjacent nodes belong to the same component and have an association relationship because they belong to the same component. Non-adjacent nodes belong to different components. The length of the trajectory from to is counted as one step, the length of the trajectory from to is counted as steps, and is called the -step reachable node of .
- 7.
The influence degree of fault propagation diffusion. The sum of the propagation and diffusion influences caused by a certain fault mode in the hypernetwork.
- (1)
Fault mode propagation influence capability.
The fault propagation influence capability of a fault mode is related to the probability of occurrence of the fault mode and the number of times it occurs in a component. According to the established hypergraph, when a certain fault mode of a component occurs many times during the operation of the machining center, it indicates that the mode of the component has a strong fault propagation capability. The function
represents the fault propagation capability of fault mode
in component
, is shown as follows:
where
.
indicates the number of times of fault mode
appearing in component
.
represents the occurrence probability of fault mode
, and the calculation equation is shown in Equation (4):
where
is the frequency of fault mode
in component
at time
.
represents the fault rate of component
at time
, and
represents the fault rate of the whole machine at time
.
According to Equation (4), combined with
Table 1 and
Table 3, the fault mode frequency ratio and fault mode occurrence probability of components at 4000 h are obtained and summarized, as shown in
Table 4.
Using the correlation properties of the correlation matrix, the propagation capability matrix
of the fault mode in each component can be calculated as
. The stronger the propagation capability of a certain fault mode, the greater the probability of causing other fault modes, and the larger the corresponding element value in matrix
.
- (2)
Calculation of fault mode propagation influence degree
According to the fault mode propagation capability matrix, the propagation influence matrix between the fault modes can be further calculated. Corresponding to the fault mode-component hypergraph, which can represent the fault propagation situation between the fault modes. The fault mode propagation influence matrix represented by the matrix
is shown in Equation (6):
where
is the transpose of matrix
and
is the diagonal matrix of
. Since the propagation of faults occurs between different modes, it is not necessary to consider the diagonal elements in
, and set it to 0. The elements in the matrix can be represented as
This matrix characterizes the fault propagation between fault modes within the same component. At this time, the fault is only propagated inside the component. The off-diagonal elements represent the influence of fault propagation between modes and , that is, the probability of causing another fault mode when one of the fault modes occurs. If the value of the off-diagonal element is 0, it means that there is no fault propagation between the fault modes.
We bring the matrix
into Equation (7) to obtain the matrix
, as shown below. The one-step fault propagation trajectory of each fault mode can be obtained from the propagation influence matrix
. Taking mode 102 as an example, its one-step propagation trajectory is
,
, and
.
- (3)
Calculation of fault mode diffusion influence degree
The fault mode propagation influence matrix
indicates the situation of fault propagation via “shared” hyperedges. In fact, the fault not only propagates once but also further diffuses. As shown in
Figure 6, the relationship between nodes and edges is
,
.
can propagate faults to
,
can propagate faults to
, and
cannot directly propagate faults to
, but
also belongs to the hyperedges
and
.
fault can still transmit the fault to
via the “springboard” effect of
. This situation is known as fault diffusion.
It should be noted that the fault probability of the
k-step reachable node is a conditional probability based on the fault of the (
k − 1) step reachable node.
In the process of fault propagation, the reliability of the component will also affect the diffusion probability of the fault mode. The higher the reliability of the component, the stronger the ability to resist the fault. Therefore, combining the conditional probability with the fault probability, the diffusion influence matrix is obtained as follows:
According to Equations (1) and (10), combined with
Table 4, the two-step fault diffusion influence matrix
can be obtained, as shown below.
We still take mode 102 as an example, whose two-step diffusion trajectory is shown in
Table 5. It is important to note that the fault will not propagate continuously. When the influence value is less than a certain value, the fault will not propagate. According to literature research [
29], this paper assumes that the fault will not diffuse when the influence degree of fault diffusion is less than 10
−8.
In theory, the two-step diffusion trajectory of mode 102 can realize fault diffusion, but via calculation, it is found that the two-step diffusion influence value of mode 102 to some nodes is less than 10
−8, as shown in the dotted line node in
Table 5. At this time, the fault will not be transmitted. Therefore, the two-step diffusion trajectory of mode 102 has a total of 8 paths reachable.
Meanwhile, according to
Table 5, it can be seen that there may be multiple diffusion paths between the two modes. For example, mode 102 can pass
and then propagate the fault to the mode 101 through
, or it can pass
and then propagate to mode 101 through
.
Similarly, the three-step fault diffusion influence matrix
of each fault mode can be calculated, as shown below.
From the matrix , it can be found that when the fault diffuses in the third step, there have been many modes with the fault diffusion influence degree less than 10−8. After the calculation of the fault diffusion in the fourth step, the fault diffusion influence degree of all modes is less than the threshold of 10−8. Therefore, the fault of the machining center occurs in three steps at most at 4000 h.
As the number of diffusion steps of fault mode increases, the diffusion influence of fault is constantly decreasing, but the influence of the comprehensive fault mode is on the rise. Therefore, although the influence of diffusion is becoming smaller and smaller, its influence on the whole fault propagation cannot be ignored.
5.3. Comprehensive Influence Degree of Fault Mode
It is assumed that after the occurrence of fault mode
,
k-step propagation and diffusion are conducted, and its influence on other nodes in the hypernetwork is expressed by the comprehensive influence degree of the fault mode:
where,
. When
k = 1, it means that the fault has one-step propagation, and only needs to calculate
. At this time, the direct propagation influence of the fault can be obtained according to the influence degree matrix of the direct propagation of the fault. The calculation equation is as follows:
When , it means that the fault has diffused further. It should be noted that when calculating , it is not necessary to re-calculate the influence of the node that has failed in the step k.
According to the multi-step propagation diffusion influence matrix, combined with the hypergraph model, the propagation diffusion influence value of each step of the fault mode can be obtained. Furthermore, the comprehensive influence value of the fault mode can be calculated by Equation (13), which is summarized in
Table 6.
As can be found from
Table 6, the modes with high comprehensive influence degree at 4000 h are 101, 503, 103, 502, 601, 603, and 803. However, the traditional methods for determining the influence of fault modes are mostly based on the subjective evaluation of humans, lacking objective quantitative indicators.