Research Article | Open Access

# A Pythagorean Fuzzy Multigranulation Probabilistic Model for Mine Ventilator Fault Diagnosis

**Academic Editor:**Marcin Mrugalski

#### Abstract

In coal mining industry, the running state of mine ventilators plays an extremely significant role for the safe and reliable operation of various industrial productions. To guarantee the better reliability, safety, and economy of mine ventilators, in view of early detection and effective fault diagnosis of mechanical faults which could prevent unscheduled downtime and minimize maintenance fees, it is imperative to construct some viable mathematical models for mine ventilator fault diagnosis. In this article, we plan to establish a data-based mine ventilator fault diagnosis method to handle situations where engineers are absent or they are incapable of coming to a conclusion from multisource data. In the process of building the mine ventilator fault diagnosis model, considering that probabilistic rough sets (PRSs) could reduce the errors triggered by incompleteness, inconsistency, and inaccuracy without needing any additional assumptions and Pythagorean fuzzy multigranulation rough sets (PF MGRSs) over the two universes’ model could effectively handle data representation, fusion, and analysis issues, we generalize the existing PF MGRSs over the two universes’ model to the PRS setting, as well as to further establish a novel model named Pythagorean fuzzy multigranulation probabilistic rough sets (PF MG-PRSs) over two universes. In the granular computing paradigm, three types of PF MG-PRSs over two universes based on the risk attitude of engineers are proposed at first. Afterwards, several basic propositions of the newly proposed model are explored. Moreover, a PF multigranulation probabilistic model for mine ventilator fault diagnosis based on PF MG-PRSs over two universes is investigated. At last, a real-world case study of dealing with a mine ventilator fault diagnosis problem is given to illustrate the practicality of the presented model, and a validity test, a sensitivity analysis, and a comparison analysis are further explored to demonstrate the effectiveness of the presented model.

#### 1. Introduction

Nowadays, the development of modern society and economy mainly depends on an adequate and reliable supply of energy to some extent. Among them, as one of the most abundant reserves of fossil fuel in the world, coal resources still play a crucial part in promoting the industrialization and social economic development of some areas; therefore, safe and reasonable exploration of coal resources is the key to most energy-dependent economies. At present, the safety production issue of coal mine is rather complicated, accidents are still in high volume in some regions, and the frequent situation of heavy and serious accidents has not been resolved radically; thus, coal mine is still an industry with high risk and a high accident rate. Specifically, mine ventilator, a rotating electrical machine which ensures the safety of coal mines and underground coal miners through providing fresh air and eliminating harmful gases, acts as an indispensable component of the coal mine safety production, energy conservation, and automated administration. In consideration of the fault detection and diagnosis of the continuous process which exert a strong influence on the product quality and production safety, it is essential to study on viable mathematical models of fault diagnosis technologies for mine ventilators [1].

In the past decades, a variety of fault diagnosis techniques have been put forward. Among them, artificial intelligence-based fault diagnosis models are recognized as a powerful scheme to enhance the effectiveness of fault diagnosis for electrical machines owing to the advent of big data era, especially in fault detection and maintenance procedures. Moreover, in various artificial intelligence-based fault diagnosis models, data-based methods have received wide attentions from scholars and practitioners since they do not need any prior knowledge about the parameters and models of electrical machines and only an information system of both normal and abnormal situations for locating faults is needed [2, 3]. Usually, neural networks, expert systems, and fuzzy approaches are common data-based fault diagnosis methods. However, the neural network-based fault diagnosis method demands compatible and plenty of training data to guarantee appropriate training, the expert system-based fault diagnosis method is hard to obtain knowledge and maintain a variable database owing to the heuristic nature of the method, and the fuzzy approach-based fault diagnosis method could avoid the above defects to some extent and cope with the complexity and uncertainty existed in mine ventilator fault diagnosis reasonably [4, 5].

During the process of real-world mine ventilator fault diagnosis, we considered that the increased volume of fault diagnosis data available to mine ventilation engineers contains numerous fuzzy, inaccurate, and inconsistent information due to the complexity and severe conditions of the coal mine environment, which is essential information for solving mine ventilator fault diagnosis problems. In addition, a fault type often implies several fuzzy, inaccurate, and inconsistent information for a certain fault characteristic, which features a relationship between fault types and fault characteristics. Hence, we work with the uncertainties to lead mine ventilation engineers to conduct a proper fault diagnosis for faulty mine ventilators. In most of the mine ventilator fault diagnosis problems, there exist some relationships between fault types and fault characteristics provided by multiple engineers, and then, engineers conduct a fault diagnosis on the basis of the similarity between an unknown sample for a faulty mine ventilator and the relationships between fault types and fault characteristics. As stated previously, fuzzy approach-based fault diagnosis methods could be seen as a reasonable model to handle such relationships and unknown samples of faulty mine ventilators.

Ever since the establishment of fuzzy approaches [6], considering that it is essential to express the uncertainty with respect to various real-world applications within a more efficient scheme that is different from classical fuzzy sets, some different types of fuzzy sets have been established one after the other during the past decades [7–10]. Among them, through applying membership degrees and nonmembership degrees simultaneously, the intuitionistic fuzzy set (IFS) theory [11] enables experts to process bipolar notions by virtue of both the epistemic certainty corresponds to membership degrees and the epistemic uncertainty corresponds to nonmembership degrees. Hence, IFSs provide a reasonable scheme to cope with numerous complete and incomplete fuzzy information in society and nature. However, it is worth noting that the sum of the membership degree and the nonmembership degree for a certain element is less than or equal to 1; experts might confront with a situation in which the sum of the membership degree and the nonmembership degree is greater than 1. In order to handle this issue, Yager [12, 13] proposed and developed the notion of PFSs, which is featured by the membership degree and the nonmembership degree as well and whose sum of squares is less than or equal to 1. Afterwards, Yager and Abbasov [14] illustrated the issue by assigning the membership degree and the nonmembership degree as and , respectively; it is noted that but ; thus, they are applicable to the context of PFSs, and the authors further pointed out that the PFS is a more reasonable and effective tool to deal with uncertain situations since PFSs could represent much more valid information than IFSs. Currently, theoretical foundations and practical applications for PFSs have become a significant theme in the area of knowledge discovery and data mining [15–24]. Hence, PFSs could better deal with an increasing number of uncertain information in mine ventilator fault diagnosis and offer mine ventilation engineers with a flexible scheme to express relationships and unknown samples of faulty mine ventilators.

Additionally, unlike existing classical information analysis approaches, due to the fact that information systems in rough sets share the identical mathematical structure with various relationships in realistic applications and the rough set theory [25–28] is aimed at handling inconsistent, incomplete, and uncertain information systems by designing lower approximations corresponding to deterministic rules and upper approximations corresponding to possible rules, it is recognized that rough set methods constitute another significant way to aid fault diagnosis technologies [29–33]. More recently, a variety of generalized rough set models have been designed in accordance with practical demands of real-world situations. Among them, the model of MGRSs over two universes [34] acts as a crucial component in solving information fusion and analysis issues for the following reasons. The first one is MGRSs over two universes take advantages of MGRSs [35, 36], which not only accelerate the process of information fusion by parallel computing multiple binary or fuzzy relations but also offer optimistic and pessimistic models corresponding to risk-seeking and risk-averse strategies; hence, MGRSs could be seen as a reasonable information fusion tool. The second advantage is that the two universes’ framework [37–45] makes up for deficiencies in representing information systems by providing some inherent relationships between the alternatives set and the attributes set, such as relationships between fault types and fault characteristics in fault diagnosis procedures. In consideration of the above two merits, studies of MGRSs over two universes-based information fusion and analysis problems are being regarded as promising research issues in recent years [19, 46–49].

In view of the importance of MGRSs over two universes in information fusion and analysis, Zhang et al. [19] constructed the notion of PF MGRSs over two universes and further investigated a general risk-based rule within the context of mergers and acquisitions. Furthermore, based on the probability theory and Bayesian procedures, it is noticed that there are families of rough set models which are equipped with the fault tolerance capability and they are robust when processing a number of noisy data [50–58]. Among them, Wong and Ziarko [50] firstly put forward the model of PRSs through combining probabilistic approximation space with rough sets. In PRSs, the corresponding rough approximations could be constructed by virtue of rough membership functions which could be seen as the probability of an arbitrary object being part of a certain set. Moreover, compared with classical rough sets, taking into account the probability of an object in a set to compute rough memberships, PRSs allow the existence of the fault tolerance by setting thresholds and , which is characterized by positive rules, negative rules, and boundary rules, respectively. From then on, through allowing a certain acceptable level of error to handle probabilistic practical applications, PRS-based research topics have received increasing attentions [59–64].

In light of the above statements, though the models of PFSs, MGRSs over two universes and PRSs are able to handle mine ventilator fault diagnosis problems by virtue of their respective features; it is noted that the absence of interaction and integrative researches on the three models precludes an effective mathematical formulation that can be utilized for mine ventilator fault diagnosis. The research gaps and motivations are listed as follows. (1)It is well known that PFSs are able to aid mine ventilation engineers to describe diverse uncertain information existed in mine ventilator fault diagnosis. Since rough set theory is a reasonable and efficient soft computing tool to cope with uncertain information systems, it is necessary to bridge the gap between PFSs and rough sets to better conduct mine ventilator fault diagnosis. Hence, we aim to develop a comprehensive rough set model within the PFS context(2)In order to discover the fault type of mine ventilators effectively, the invitation of several mine ventilation engineers to conduct a group decision-making outperforms an individual decision-making. Thus, it is necessary to bridge the gap between PF rough sets and group decision-making models. Considering the merits owned by MGRSs over two universes when making group decisions, we aim to investigate a novel MGRSs over the two universes’ model within the PFS context(3)In view of the importance of fault tolerance capability and robustness owned by the mine ventilator fault diagnosis approach, which further avoids the influence of decision errors, according to the advantages of PRSs mentioned above, it is necessary to bridge the gap between PF MGRSs over two universes and PRSs. Thus, in order to let the newly proposed model manifests robustness in mine ventilator fault diagnosis processes, as well as to expand theoretical scopes of PRSs and MGRSs over two universes, we intend to put forward the model of PF MG-PRSs over two universes by integrating PF MGRSs over two universes with PRSs and three types of the proposed model are explored. Moreover, on the basis of PF MG-PRSs over two universes, a general PF multigranulation probabilistic model within the mine ventilator fault diagnosis context is further discussed. Finally, a case study concerning a mine ventilator fault diagnosis issue is explored to show the validation of the proposed method

Comparing with existing models to mine ventilator fault diagnosis, the primary contributions of the article are listed as follows. (1)In light of the granular computing paradigm in the field of artificial intelligence-based information processing, we put forward the notion of PF MG-PRSs over two universes to handle mine ventilator fault diagnosis, which could be seen as a beneficial supplement for artificial intelligence-based fault diagnosis models. Additionally, we explore several common propositions of the proposed model(2)The proposed model takes advantages of PF MGRSs over two universes and PRSs simultaneously. The first merit is that the utilization of PF MGRSs over two universes enables engineers to fuse information based on parallel computing and the risk attitude of engineers and the variable type of PF MG-PRSs over two universes could overcome limitations of optimistic and pessimistic counterparts for coping with variable risk attitudes in fault diagnosis. Moreover, the second merit is that the utilization of PRSs makes the proposed model owns the capacity of processing a number of noisy data, which is featured by positive, negative, and boundary fault diagnosis rules(3)By virtue of variable PF MG-PRSs over two universes, a PF multigranulation probabilistic model for mine ventilator fault diagnosis with PFSs is constructed and a numerical example, a validity test, a sensitivity analysis, and a comparison analysis illustrate that the superiorities of the proposed fault diagnosis rule could cut down uncertainty and improve accuracy of mine ventilator fault diagnosis to a great extent

The rest of this article is set out below. The next section concisely revisits several preliminary backgrounds about PFSs, PF MGRSs over two universes, and PRSs. In Section 3, three types of PF MG-PRSs over two universes are established, i.e., optimistic, pessimistic, and variable PF MG-PRSs over two universes and then, several fundamental propositions of them are also concluded. Section 4 constructs a general PF multigranulation probabilistic model based on PF MG-PRSs over two universes. In Section 5, an illustrative real-world example concerning a mine ventilator fault diagnosis problem is provided to show the practicability of the proposed model. At last, a few conclusive comments and future study topics are included in Section 6.

#### 2. Preliminaries

For the convenience of the following statements, we present some fundamental knowledge about PFSs, PF MGRSs over two universes, and PRSs.

##### 2.1. PFSs

In [12, 13], Yager proposed the notion of PFSs, which is a generalization of IFSs [11] and the mathematical form of IFSs and PFSs could be presented below.

*Definition 1 (see [11]). *Suppose is an arbitrary nonempty finite universe of discourse, an IFS in , represented by , is given by
where and represent the membership degree and the nonmembership degree of the element to , respectively. For all , it holds that . Moreover, the degree of indeterminacy is denoted as .

*Definition 2 (see [12, 13]). *Suppose is an arbitrary nonempty finite universe of discourse, a PFS in , represented by , is given by
where and represent the membership degree and the nonmembership degree of the element to , respectively. For all , it holds that . Moreover, the degree of indeterminacy is denoted as . For convenience, is named a Pythagorean fuzzy number (PFN) by Zhang and Xu [16] that satisfies and . In addition, we call the family of all PFSs in as .

Next, given two PFSs denoted by and , then some operations for and were presented in [16, 17].

*Definition 3 (see [16, 17]). *Suppose , . For all , we have

To investigate the magnitude of different PFNs, the comparison law by virtue of score functions and accuracy functions is presented as follows [16, 17].

*Definition 4 (see [16, 17]). *Suppose and are two PFNs, and let and be their score functions and and be their accuracy functions, then
(i)If , then (ii)If , then(1), then (2), then

##### 2.2. PF MGRSs over Two Universes

In view of the superiority of MGRSs over two universes in information fusion and analysis situations, Zhang et al. [19] proposed PF MGRSs over the two universes’ model. Before introducing such a model, it is essential to develop the notion of PF relations over two universes.

*Definition 5 (see [19]). *Suppose and are two arbitrary nonempty finite universes of discourse. A PF relation is a PFS in , i.e., is given by
where and . For all , it holds that . In addition, we call the family of all PF relations over two universes in as .

On the basis of PF relations over two universes discussed above, the notion of PF MGRSs over two universes could be established according to the constructive approach.

*Definition 6 (see [19]). *Suppose and are two arbitrary nonempty finite universes of discourse, and . Then, is named a PF multigranulation approximation space over two universes. For any , the optimistic PF lower and upper approximations of in terms of , represented by and , are two PFSs given by
where
In an identical fashion, the pessimistic PF lower and upper approximations of in terms of , represented by and , are also two PFSs given by
where
In light of the above statements, we denote and the optimistic and pessimistic PF MGRSs over two universes, respectively.

##### 2.3. PRSs

In classical rough sets, the approximation space is exactly deterministic since rough sets are based on a deterministic knowledge base; hence, this potential weakness may preclude classical rough sets from including some objects into the approximation region. In order to conquer this limitation, some fundamental notions of PRSs are revisited in a brief way [50].

Suppose is an arbitrary nonempty finite universes of discourse, is an equivalence relation in , is a probability measure on the basis of the -algebra that is combined by the subset family of . Then, we denote the triple as a probabilistic approximation space.

*Definition 7 (see [50, 54, 55]). *For any and , the lower and upper approximations of with respect to , represented by and , are given by
where represents the probability of the object in given that the object is in . Moreover, the corresponding positive region, negative region, and boundary region in terms of the probabilistic lower and upper approximations are represented below.

It is noted that with the parameter decreases, the success rate of an object is correctly classified decreases, while if the value of the parameter is large, the success rate of an object is correctly classified decreases as well. In particular, PRSs reduce to classical rough sets when and .

#### 3. PF MG-PRSs over the Two Universes’ Model

In the following section, we generalize the model of PF MGRSs over two universes to the context of PRSs and further develop three types of PF MG-PRSs over two universes, i.e., the model of optimistic, pessimistic, and variable PF MG-PRSs over two universes; then, some of their fundamental propositions will be discussed briefly.

##### 3.1. Optimistic PF MG-PRSs over the Two Universes’ Model

According to the original definition of optimistic MGRSs, the reason for using the word “optimistic” lies in at least one granular structure is needed to meet the demand of the inclusion condition between an equivalence class and the approximated target when computing the optimistic multigranulation lower approximation, while the optimistic upper approximation within the multigranulation context is defined based on the complement of the optimistic multigranulation lower approximation.

Following the idea, in the background of PRSs, the corresponding optimistic multigranulation lower approximation should also utilize at least one granular structure to meet the demand of the probability constraint between an equivalence class and the approximated target, while the corresponding optimistic multigranulation upper approximation should be computed in a similar manner.

Next, the concept of PF inclusion degrees, which denotes membership degrees of some objects in corresponding PFSs, is given as follows.

*Definition 8. *Suppose and are two arbitrary nonempty finite universes of discourse, is a PF relation in . For any , , and , the membership degree of in in terms of , denoted by , is given as

It is worth noting that is constructed based on a single PF relation over two universes. However, the multigranulation context requires experts to utilize multiple PF relations over two universes when computing the corresponding multigranulation lower and upper approximations. Hence, we give the following definition by considering multiple membership degrees.

*Definition 9. *Suppose and are two arbitrary nonempty finite universes of discourse, is a PF relation in . For an*y *, , and , the maximal and minimal membership degrees of in in terms of each , denoted by and , are given as

By means of the concept of maximal membership degrees , optimistic PF MG-PRSs over two universes could be designed below.

*Definition 10. *Suppose and are two arbitrary nonempty finite universes of discourse, is a PF relation in . For any , , and , threshold parameters and are two PFNs and ; the optimistic PF multigranulation probabilistic lower and upper approximations of in terms of , denoted by and , are given as
The pair is called an optimistic PF MG-PRS over two universes. Moreover, the corresponding positive region, negative region, and boundary region in terms of the optimistic PF multigranulation probabilistic lower and upper approximations are represented below.

*Example 1. *Let and be two universes, , and be three PF relations in , where

Then, a PFS in is given as follows.

According to Definition 8, we obtain

In an identical fashion, we also obtain , , , , , , , , , , and .

Then, according to Definition 9, we obtain , , , and .

If we take and , we obtain and . Then, , , and could be obtained.

Proposition 1. *Given a PF multigranulation approximation space over two universes . For any , threshold parameters and are two PFNs and ; then, the optimistic PF multigranulation probabilistic lower and upper approximations have the following properties.
*

*Proof. *(1)According to Definition 10, since , we haveHence, could be obtained easily.
(2)According to Definition 10, we haveHence, we obtain , and can be obtained in an identical fashion.
(3)In view of the above conclusions, and can be proved similarly.(4)According to Definition 10, since , we haveHence, we obtain . Similarly, could be obtained easily.

##### 3.2. Pessimistic PF MG-PRSs over the Two Universes’ Model

Contrary to the idea of optimistic MGRSs, according to the original definition of pessimistic MGRSs, the word “pessimistic” indicates that experts utilize all granular structures to meet the demand of the inclusion condition between an equivalence class and the approximated target when computing the pessimistic multigranulation lower approximation, while the pessimistic multigranulation upper approximation is also defined on the basis of the complement of the pessimistic multigranulation lower approximation.

In order to expand the model of pessimistic PF MGRSs over two universes to the PRS setting, the corresponding pessimistic multigranulation lower approximation should be constructed based on all granular structures to meet the demand of the probability constraint between an equivalence class and the approximated target, while the corresponding pessimistic multigranulation upper approximation should also be computed in an identical fashion.

Similar to optimistic PF MG-PRSs over two universes, according to the concept of minimal membership degrees , we present the notion of pessimistic PF MG-PRSs over two universes below.

*Definition 11. *Suppose and are two arbitrary nonempty finite universes of discourse, is a PF relation in . For any , , and , threshold parameters and are two PFNs and ; the pessimistic PF multigranulation probabilistic lower and upper approximations of in terms of , denoted by and , are given as
The pair is called a pessimistic PF MG-PRS over two universes. Moreover, the corresponding positive region, negative region, and boundary region in terms of the pessimistic PF multigranulation probabilistic lower and upper approximations are represented below.

*Example 2. *Given a PF multigranulation approximation space over two universes which is defined in Example 1. According to Definition 9, we obtain , , , and .

If we also take and , we obtain and . Then, , , and could be obtained.

Proposition 2. *Given a PF multigranulation approximation space over two universes . For any , threshold parameters and are two PFNs and ; then, the pessimistic PF multigranulation probabilistic lower and upper approximations have the following properties.
*

##### 3.3. Variable PF MG-PRSs over the Two Universes’ Model

From the standpoint of classical optimistic and pessimistic MGRSs, it is worth noting that optimistic and pessimistic MGRSs are established by using at least one granular structure and all granular structures, respectively, which act as two extreme models in the procedure of information fusion and they have no capacity to adjust risks. Furthermore, it is also noted that optimistic and pessimistic PF MG-PRSs over two universes are constructed by virtue of maximal and minimal membership degrees and they lack the adjustable capability as well. Thus, optimistic and pessimistic PF MG-PRSs over two universes are fixed and could not be changed according to the risk preference of experts. In the following, through introducing a risk coefficient , the concept of variable membership degrees based on maximal and minimal membership degrees could be put forward and variable PF MG-PRSs over two universes could be further developed.

Next, we give the following definition by combining risk coefficients with maximal and minimal membership degrees.

*Definition 12. *Suppose and are two arbitrary nonempty finite universes of discourse, is a PF relation in . For any , , and , the variable membership degree of in in terms of each , denoted by , is given as

It is noted that the variable membership degree reduces to the maximal membership degree when , while the variable membership degree reduces to the minimal membership degree when . On the basis of variable membership degrees , we propose the concept of variable PF MG-PRSs over two universes as follows.

*Definition 13. *Suppose and are two arbitrary nonempty finite universes of discourse, is a PF relation in . For any , , and , threshold parameters and are two PFNs and ; the variable PF multigranulation probabilistic lower and upper approximations of in terms of , denoted by and , are given as
The pair is called a variable PF MG-PRS over two universes. It is easy to see variable PF MG-PRSs over two universes reduce to the optimistic counterpart and pessimistic counterpart when and , respectively. Thus, optimistic and pessimistic PF MG-PRSs over two universes could be regarded as special cases of variable PF MG-PRSs over two universes. Moreover, the corresponding positive region, negative region, and boundary region in terms of the variable PF multigranulation probabilistic lower and upper approximations are represented below.

*Example 3. *Given a PF multigranulation approximation space over two universes which is defined in Example 1. According to Definition 12, if we take the risk coefficient , it is not difficult to obtain , , , and .

If we also take and , we obtain and . Then, , , and could be obtained.

Proposition 3. *Given a PF multigranulation approximation space over two universes . For any , threshold parameters and are two PFNs and ; then, the variable PF multigranulation probabilistic lower and upper approximations have the following properties.
*