Development of Excessive-Smoke-Emitting-Car Fault Diagnostic System (ESECFDS) based on Case-Base Reasoning (CBR) Methodology

Abstract

This paper proposes a diagnostic system, ESECFDS, to solve the problem of excessive smoke emission from a faulty car, based on Case-Based Reasoning (CBR) Methodology that accepts a user query as a new case, compares it with several stored cases in a Case-Base (CB) based on a similarity score using Jaccard Similarity method and provides feedback result to the user. The CBR cycle R4 Model of Aamodt and Plaza has been modified to R5 Model to improve the time complexity of the proposed algorithm of the system over the popular R4 Model.

Keywords

Excessive-Smoke-Emitting-Car Fault Diagnostic system ESECFDS CBR Case Decision Tree Sub-Case-Base Jaccard Similarity Method User Query Feedback

1. Introduction

Since the 20th century, the importance of the car in daily life as a means of conveyance has been discussed in [1]. The huge growth of car usage has a visible significant impact on our society. But many drivers face difficulties when their cars become faulty and out of control especially in transit. Detection of car faults is a complex process which involves a high degree of skill to identify and fix the fault within the specified time. An expert mechanic or driver, who has a lot of knowledge of finding fault, can do the same. Experts deal with various levels of difficulties of car fault; some being detected with visible effects. One of them is Excessive-Smoke-Emitting-Car. Sometimes, a car may be poorly tuned or maintained; it may emit an excess level of smoke. Smoke is a cause of concern and may suggest several problems that may happen underneath the hood of the car. Smoke colour may be black or white or blue. So, different colours of smoke coming out of the exhaust pipe may indicate different serious problems in the exhaust system, engine, or inside the fuel injection system. If expert knowledge can be documented in a computer to solve problems, dependence on the experts may be minimized.

This paper proposes the Excessive-Smoke-Emitting-Car Fault Diagnostic System (ESECFDS) that solves the problems of a faulty car that is emitting excessive smoke from the exhaust pipe. Using computational intelligence, ESECFDS determines the solution to the fault. The system, developed in this paper, is based on the principle of Case-Based reasoning. Case-Based Reasoning (CBR) is a technique for computer-aided intelligent diagnosis system. ESECFDS provides a query format to the user, so that the user may provide a new query case to the system. The case takes in a few attributes that uniquely identify the case along with the problem description & solution. The case is represented as Case $=$ (problem description, solution). Problem is represented as problem description $=$ case attributes (feature, values). CBR methodology-based ESECFDS retrieves the most similar cases from previously stored cases in the Case-Base (CB) and responds with solutions to the user. This paper also focuses on minimizing the execution time of fetching solution from the Case-Base. To minimize the execution time of the fetching solution, cases are repartitioned within the Case-Base and stored into some Sub-Case-Bases. The system uses a decision tree at the time of the cases are stored in the different Sub-Case-Bases. The cases with the same smoke-colour value are stored into the same Sub-Case-Base of the Case-Base. For example, the cases with an attribute-value of smoke-colour value $=$ “Black”, are to be stored in the same Sub-Case-Base, similarly cases with smoke-colour value $=$ “White”, are to be stored into another Sub-Case-Base.

ESECFDS uses the Jaccard Similarity Method for the estimation of similarity between the new case and the stored cases. Jaccard similarity method is very simple and performs calculations between a pair of sets in minimum time. Section-9 shows the detail about the performance of the proposed methodology with the Jaccard similarity method.

The future work focuses on the possibility of designing of diagnosis system for car faults based on CBR Methodology that will deal with all types of car problems.

The remainder of the paper is organized as follows: Section 2 focuses on Case-Based Reasoning (CBR) Methodology; Section 3 on previous related works, Section 4 depicts the comparative analysis of prior works and R4, R5 model in ESECFDS. Section 5 focuses on proposed model architecture, Section 6 provides detail on Case-Base and case representation in the Case-Base, Section 7 describes our proposed methodology for similarity calculation between cases, and Section 8 describes the proposed algorithm in ESECFDS with simulation examples. Section 9 focuses on the performance analysis of proposed methodology in ESECFDS, Section 10 focuses on the limitations of ESECFDS and at last Section 11 ends up with a conclusion and future work.

2. Case-Base Reasoning Methodology

A general Artificial Intelligence methodology used for problem-solution, fault identification and treatment, learning, reasoning, and decision support is Case-Based reasoning or CBR. The technique of CBR is the concept of solving a new issue using old issues, so it is often called the approach of problem-solution based on experience. It is a technique for intelligent systems of computer-aided diagnosis. A case includes a description of the problem (i.e. features or symptoms) and a solution (i.e., a result of diagnosis). A Case-Base consists of cases those are previously stored. As seen in [2], many similar cases can have similar solutions. By searching for similar problems in the Case-Base, a new problem is solved. If with the same problem a similar case is found, then the solution to that problem will be adapted to be the solution to the new problem. When a new solution to the new problem or case has been found then that case with a new solution can be stored in the Case-Base in order to improve its competence, as referred to in [2]. CBR works in a cycle. Aamodt and Plaza in 1994, proposed the CBR cyclic model called the R4 model that has 4 activities; Retrieve cases, Reuse Cases, Revise cases, and Retain Cases. This paper used the R5 model of CBR. As in [3] shows, the R5 model provides repartitioning of the case and case structure before retrieval that may reduce search time for queries.

There are 5 activities of the R5 model.

1.
Repartitioning: R5 models are partitioning the cases with the help of the content of its collection of cases as referred to in [3]. It’s an important task to build the structure of cases before retrieving the case.
2.
Retrieve: Repartitioning provides a foundation for the retrieve case. Now in the retrieving stage, the case is retrieved from the closest similar group of cases. With the help of a similarity relationship, first is chosen as the most similar group. Then among the cases in a similar group, the closest case is chosen, as referred to in [3].
3.
Reuse: This activity follows that the concept of a similar problem has a similar solution. R5 model solves the new problem or case re-using the solutions, contained in those stored retrieved case(s).
4.
Revise: After the step of reuse, the R5 model verifies the applicability of the newly proposed solution in the real world.
5.
Retain: At last, a new solved problem is updated to the Case-Base for future problem solution and finds the position in the most similar group of a case in case library as referred to in [3]. Figure 1 shows the CBR cycle of R5 Model.

Figure 1.
CBR cycle of R5 Model.

2.1 CBR in ESECFDS

Excessive-Smoke-Emitting-Car Fault Diagnosis System (ESECFDS) based on Case-Based Reasoning Methodology, follows the activities of the R5 Model of CBR.

1.
Repartitioning: The Case-Base of ESECFDS stores cases which contain excessive-smoke- emission problems with solutions. As car may emit different colours of excessive smoke so cases may contain problems of smoke emission in different colours. So, according to the different smoke colours, Case-Base (CB) is repartitioned into different Sub-Case-Bases (SCB) where cases with same smoke colour are stored into the same Sub-Case-Base.
2.
Retrieve: After the repartition stage, whenever a problem or a new case arrives in the system, the most similar case is retrieved from the closest similar Sub-Case-Base. With the help of the decision tree first is chosen as the most similar Sub-Case-Base. Then among the cases in that Sub-Case-Base, the most similar case is chosen.
3.
Reuse: This activity follows the concept of a similar problem has a similar solution. So, at this stage, the problem of the new case is solved re-using the solutions of the retrieved case.
4.
Revise: After the step of reuse, verification of the applicability of the newly proposed solution in the real world is done.
5.
Retain: At last, the newly solved case is updated to the most appropriate Sub-Case-Base for future problem-solution.

3. Previous related works

In various diagnostic fields, the expert systems are used and can be well applied in the diagnosis of automobile faults.

A Vehicle Fault Diagnosis Expert System based on the fault tree analysis method was developed, as referred to in [4], to solve the complexity and difficulty of detection and analysis of vehicle faults. That designed expert system can effectively diagnose vehicle breakdowns and help users to exclude faults and maintain vehicles.

Car Failure and Malfunction Diagnosis Assistance System (CFMDAS) has been proposed in [5], using a decision tree which is extremely useful for supporting mechanics for the identification of car failure and training purposes. In addition, more specific and less time-consuming models are developed by collecting and maintaining useful information on such domains.

A distributed diagnosis agent system (DDAS) was built in [6], which is used on the basis of signal analysis and machine learning for vehicle fault diagnosis. Based on the information provided by the signal agents in DDAS, DDAS used CBR techniques to find the root cause of vehicle faults.

As proposed in [7], researchers have built a car malfunction fuzzy fault diagnostic system based on fault tree analysis that detects component faults and process disturbances. It can lead to system malfunctions by matching the process uncertainty data from the system with the pattern of IF statements stored in the computers. Surety factors for component failures and process disturbances for diagnostic sequence checking are also evaluated by that system.

A method of vehicle fault diagnosis that combines neural networks with generalization capability is proposed in the research paper [8], where a two-step ensemble strategy was proposed for the diagnosis of vehicle faults, an ensemble selection algorithm, BFES, and an analog Bayesian ensemble decision function, A-Bayesian-Entropy.

For their diagnostics, several papers have used rough set theory as in [9] to find fault in the vehicular transmission system, the rough set theory is used to deal with vehicle transmission system fault data.

Paper [10] gives an example of diesel engine valve clearance fault diagnosis and describes the algorithms and methods for extracting fault symptoms based on the principle of rough sets.

The power transformer fault diagnosis based on association rules obtained by the rough set has been seen in [11]. Here, a rough set is provided to produce rules of association that are used to diagnose power transformer faults.

In Semantic Analysis of Natural Language Queries Using Domain Ontology for Knowledge Access from the Database, the Jaccard similarity approach was used in [12], this paper describes a method for semantic analysis of natural language queries for Natural Language Interface to Database (NLIDB) using domain ontology. It needs better precision to incorporate NLIDB for extreme applications such as railway investigation, airway investigation, corporate, or government call centers.

The research paper [13] was done to help build an expert system for the diagnosis and repair of car failures under constraints such as time, location and availability of human expertise. In order to conclude the best means, which are clear and straightforward, to introduce and maintain when creating an expert system, the analysis of technologies for designing expert systems was undertaken.

Taking the improved nearest neighbor method which is based on the combination of hamming distance and Euclidian distance to solve case similarity, improve accuracy and efficiency of case matching, the research paper [14] shows the CBR retrieval method based on rough set theory.

Case-Based reasoning is used in the research paper [15] to help fault diagnosis using sensor information. This paper establishes that, based on sensor information and using Case-Based reasoning methods, computer-based diagnostic systems can be developed.

This paper [16] focuses on the application of CBR in a manufacturing environment with decision tree induction to examine the cause of defects occurring in the domain. By combining CBR with decision trees, the abstraction of domain information is made possible.

The diagnosis of automobile engine faults using the neural network was mentioned in [17]. The developed system is based on an engine fault table. Such a diagnostic module is intended to improve the system ’s usefulness.

In paper [18], a decision tree is applied for fault detection and classification in the transmission line. The DT based fault detection algorithm uses 1/4th cycle data of fault currents from fault inception, to generate the optimal decision tree for fault detection

The paper referred to in [19] provides an overview to the concept of rough sets as well as its benefit in working with uncertain knowledge in areas of defect detection. Here, a weight determination algorithm based on rough sets is proposed.

In [20], an effective random decision tree algorithm was discussed for Case-Based reasoning systems. To generate a stronger hybrid algorithm, this paper combines this algorithm with a simple similarity measure based on domain knowledge. This hybrid algorithm generates a lower average error consistently than the base algorithms

Reference was made to the development of a fault-diagnosis method (FDS) using Case-Based reasoning in [21]. Before the event of total system failure, the Fault-diagnosis system (FDS) could be helpful in determining the source of faults. Here is a study on such a diagnostic technique for minimizing maintenance activities and reducing downtime of the system.

For computer fault diagnosis, Case-Based reasoning is used in paper [22]. This system is based on the Case-Based Reasoning (CBR) methodology where the problem-solution depends on case, and the knowledge base contains certain cases. The system gives users the right to send questions and problems.

As detailed in [23], Case-Based reasoning is extended to complex medical diagnoses. This paper presents a CBR expert method that uses the algorithm of the K-nearest neighbor (KNN) to look for related cases based on the Euclidean distance calculation.

4. Comparative analysis

4.1 Comparative analysis of previous works with their outcomes

The previous related research papers used different methods for different purposes. Table 1 shows a comparative study of these works with their outcomes.

Table 1
Comparative study of previous related works with their outcomes

Previous related work	Objective of the work	Method used	Outcome analysis of the work
Referred to in [4]	Development of expert system that can diagnose vehicle failures appropriately and enable users to point out faults and maintain vehicles.	Fault tree analysis	This expert system analyses the condition of the vehicle, carefully considers every component, reasons and provides the users with potential results that can be selected by careful consideration. Even if any fault information is missing from the system, it can also produce diagnostic results.
Referred to in [5]	The research’s main objective is to develop a car failure and malfunction diagnosis assistance system (CFMDAS).	Decision tree	Outcomes of system implementation suggest that the system has value in areas where efficiency of work is increased by decreasing time for fault detection and the comprehension of the users. Further enhancements to the requirements of the system domain knowledge are needed to enhance the representation of domain knowledge. And the implementation of another AI technique to work in the revision of system rules would be considered to improve the effectiveness of the diagnosis process.
Referred to in [6]	Development of distributed diagnosis agent system (DDAS) that used for vehicle fault diagnosis.	Signal analysis and nearest neighbor retrieval method, feature matching	CBR techniques are developed for signal fault analyses within the framework of a distributed agent system. the techniques used to the diagnosis of vehicle faults and the results of the experiments have shown that both methods are very successful.
Referred to in [7]	Development of car failure fuzzy fault diagnostic system that identifies component failures and process disturbances.	Fault tree analysis	The fuzzy fault diagnostic system can detect device faults and process disruptions effectively that may result in system malfunctions by matching system process instability data with the pattern of IF statements stored in the computers. The system also evaluates security factors of component failures and process disruptions for sequence testing in diagnosis from the uncertainty observed data and information.
Referred to in [8]	Proposes a process of vehicle fault diagnosis that ensemble neural networks with generalization capability. It approaches a two-step ensemble strategies for vehicle fault diagnostics.	An ensemble selection algorithm, BFES, and two trainable decision functions, BP-Bayesian and A-Bayesian Entropy	The experimental results show that over all decision functions, our ensemble selection algorithm BFES performs better than the Top-k process, while the BFES ensemble combined with the decision function, A-Bayesian-Entropy gives the best generalization capability.
Referred to in [9]	Process the fault data of the vehicle transmission system and diagnose the fault.	Rough set theory	The rough set theory minimizes the vector character and analyses the consistency of decisions. And the outcome shows that the theory is effective in identifying gearbox faults, based on a large number of tests.
Referred to in [10]	Study on symptoms of vehicle fault extraction based on rough set theory in mechanical engineering.	Rough sets theory	The reduct algorithm of rough sets theory will help define the relationship between some main factors and specific locations of faults quickly and extract the symptoms of faults, thereby achieving the goals of sorting out and selecting the necessary. The conclusion shows the possibility of using combined application of fuzzy theory and neural network theory to establish intelligent fault diagnosis.
Referred to in [11]	Usage of rough set to produce association rules that are used for power transformer fault diagnostics.	Rough set theory	The results of the tests indicate high diagnostic accuracy of the association rules obtained by rough set.
Referred to in [12]	Design of semantic analyzer for railway inquiry domain.	Natural language queries, Jaccard similarity method	A semantic analyzer is built using English query language modelling which analyses the English language query being pre-processed to interpret its intended meaning. In the ontological form of the domain the concept is interpreted after interpretation. For the target database, this intermediate type of meaning can be generated to corresponding database query.

Previous related work	Objective of the work	Method used	Outcome analysis of the work
Referred to in [13]	Assist in the creation of an expert system for diagnosing and fixing car failures under constraint, such as time, location and availability of expert knowledge.	Rule based method	Study of technologies for designing expert systems was undertaken to conclude best means, which are simple and easy, to implement and maintain while developing an expert system.
Referred to in [14]	Analysis of CBR retrieval method focused on the rough set theory.	Combining Hamming distance and Euclidean distance	Rough set theory to assess the weight of attributes in the CBR system will prevent arbitrary weight distribution affecting the outcome of the retrieval. Improved the distance from nearest-neighbor system function. The combination of Hamming distance and Euclidean distance solves the issue of qualitative attribute similarity calculation. This approach increases the performance and consistency of CBR retrieval as opposed to the conventional approach of CBR retrieval.
Referred to in [15]	Case-Based reasoning allows the diagnosis of fault using sensor information.	Sensor information, machine learning techniques	This paper establishes that, based on sensor information and using Case-Based reasoning methods, computer-based diagnostic systems can be developed.
Referred to in [16]	Describe the application of CBR with the implementation of a decision tree in a manufacturing environment to examine the cause of the domain defects.	Decision tree	Decision tree learned to distinguish defective cases from non-defective cases can be used to classify attributes the importance of which may be the source of defects.
Referred to in [17]	Develop of automobile engine fault diagnosis using neural network.	Neural network	The neural network output is obtained in two ways. Output explicitly shows faults for the clear faults, while output is obtained in terms of labels for the incipient faults. The approach to detecting faults would be useful in discovering faults early.
Referred to in [18]	Apply decision tree for fault detection and classification in transmission line.	Decision tree	The results demonstrated that the proposed approach can accurately identify and distinguish transmission line failures within the time frame of the sub-cycle while maintaining the computational burden low enough to make a real-time implementation feasible.
Referred to in [19]	Develop a new method of Case-Based reasoning based on rough sets. An algorithm for weight determination is proposed, based on rough sets.	Rough set	Case-Based reasoning makes good use of previous experience which can help solve many practical problems in the fields of industry. Experimental results on fault diagnosis using Centrifugal pump’s fault Case-Base shows a very good accord
Referred to in [20]	Propose an Efficient Random Decision Tree Algorithm for Case-Based Reasoning Systems. Construct a stronger hybrid algorithm, combine this algorithm with a simple similarity measure based on the domain information.	Random decision tree	Test the resulting algorithms in the palliative care domain on a case basis with patient data. Our hybrid algorithm generates a significantly lower average error than the base algorithms.
Referred to in [21]	Development of web-based FDS for an injection moulding machine (IMM) on the basis of Case-Based reasoning (CBR) from previous maintenance experience.	Nearest neighbours	The resulting FDS achieves good performance, with the ability to support remote management of IMM.
Referred to in [22]	Apply Case-Based reasoning for computer fault diagnosis.	Prototype methodology list matching	It is possible to reduce the issue of lack of information technology (IT), and users can gain more expertise and experience from this method.

Table 1, continued

Previous related work	Objective of the work	Method used	Outcome analysis of the work
Referred to in [23]	Proposes a CBR-based expert system for complex medical diagnosis that uses the K-nearest neighbour (KNN) algorithm to search k similar cases based on the Euclidean distance measure.	K-nearest neighbour (KNN)	The tests have shown that the established tool works well, and the medical doctors support its level of diagnostic accuracy.

Table 1, continued

4.2 Comparative analysis of previous works those used Case-Based reasoning methodology in automobile domain

Comparative analysis of previous works those used Case-Based Reasoning Methodology in Automobile Domain is shown in Table 2.

Table 2
Comparative previous work analysis of various Case-Based Reasoning Methodologies used in automobile domain

SL No.	Authors and year	Work	CBR Methodology	CBR model	Case retrieval method	Work abstract
1.	Authors: Fan Zhang, Zude Zhou, and Xiaojie Liu Year: 2012	Research of automobile quality fault diagnosis method based on CBR	CBR quality fault diagnosis method used CBR reasoning process that consisted by four parts retrieve, reuse, revise and retain	R4 model	The discrete matching and multi-level semantic matching algorithms	In this work [24], Case-Based reasoning, theory is applied to the automobile quality fault diagnosis field. Case description, case retrieve and case reuse are the main factors of CBR. It is realized by the methods of fault tree construction, discrete cases information entropy weight value calculation, discrete and multi-level semantic matching case retrieval and knowledge difference driving case reuse. The process of solving actual cases verifies the validity and efficiency of CBR method in the quality fault diagnosis system.
2.	Authors: Ziyan, Wen, Jacob Crossman, Yi Murphey Year: 2003	Case-Base reasoning in vehicle fault diagnostics	This research work focuses on case representation, case retrieval, case reuse, case revise and case retain	R5 model	Nearest neighbor retrieval method, feature matching	This paper [6], presents research in Case-Based reasoning (CBR) with application to vehicle fault diagnosis. Here, a distributed diagnostic agent system, DDAS is developed that detects faults of a device based on signal analysis and machine learning. The CBR techniques are used to find root cause of vehicle faults based on the information provided by the signal agents in DDAS. Two CBR methods are presented, one used directly the diagnostic output from the signal agents and another uses the signal segment features
3.	Author: Diana M.L. Wong Year: 2011	Integrated reasoning approach for car faulty diagnosis	Here CBR process involves four main activities: Retrieve, reuse, revise, and retain.	R4 model	K-nearest neighbors’ algorithm	Here in [25], an integrated Case-Based and rule based reasoning method for car faulty diagnosis. The reasoning method is done through extracting the past cases from the Proton Service Center while comparing with the preset rules to deduce a diagnosis/solution to a car service case. New cases will be stored to the knowledge base. The test cases examples illustrate the effectiveness of the proposed integrated reasoning.
4.	Author: Stefan Vacek, Tobias Gindele, J. Marius Zollner and Rudiger Dillmann Year: 2007	Situation classification for cognitive automobiles using Case-Based reasoning	Case-Based reasoning for the interpretation of traffic situations is: Create case, retrieve cases, reuse cases, revise cases and retain cases.	R4 model	Case matching by traversing the Case-Base recursively along the paths given by the hierarchical organization	[26] Shows driving a car in urban areas autonomously requires the ability of an in-depth analysis of the current situation. For understanding the current situation and deducing consequences for the execution of behaviors (maneuvers), higher-level reasoning about the situation has to take place. In this paper, an approach for situation interpretation for cognitive automobiles is presented. The approach relies on Case-Based reasoning to predict the evolvement of the current situation and to select the appropriate behavior.

SL No.	Authors and year	Work	CBR Methodology	CBR model	Case retrieval method	Work abstract
5.	Author: Stefan Vacek, Tobias Gindele, J. Marius Zollner and Rudiger Dillmann Year: 2007	Using Case-Base reasoning for autonomousvehicle guidance	The order of tasks for Case-Base reasoning for the interpretation of traffic situations is: Create case, retrieve cases, reuse cases, and retain cases.	R4 model	Case matching by traversing the Case-Base recursively along the paths given by the hierarchical organization	In vehicle guidance in complex scenarios such as inner-city traffic requires an in-depth understanding of the current situation. In order to select the appropriate behavior for an autonomous vehicle, an analysis of the situation is needed. The analysis consists of an estimation of the situation’s development with respect to the selected behavior. This can only be done using higher-level reasoning techniques. In this paper, an approach for situation interpretation for autonomous vehicles is presented. The approach relies on Case-Based reasoning in order to predict the evolvement of the current situation and to select the appropriate behavior.

Table 2, continued

4.3 Comparison analysis of CBR R4 model and R5 model in ESECFDS

Comparison analysis of the CBR R4 Model and R5 Model in ESECFDS is shown in Table 3.

The Fig. 2 graph shows the comparison of total calculation time between R4 and R5 Model in ESECFDS.

Table 3
Comparison analysis of CBR R4 model and R5 model in ESECFDS

R4 model in ESECFDS

R5 model in ESECFDS

CBR R4 model does not repartition the cases before retrieval. As a result whenever a new problem with a particular smoke colour emission arrives in the system then that case shall be compared with all cases stored in the Case-Base. For example, if a new case contains smoke colour is black, and then similarity calculation will be done between the new case and all cases in the CB which may contain smoke colour grey, blue, and white. For this reason, searching the solution may take long time.

CBR R5 model follows the repartition stage, so with R5 model searching time may reduce as the new case shall be compared with some of the cases that are stored in one Sub-Case-Base.

Let, Case-Base holds total ‘

M

’ number of cases and each case contains ‘

N

’ number of attributes. So, when new query case arrives, comparison is done between each attribute-value of new query case with each of corresponding attribute-value of each stored cases in CB. If comparing with

N

attributes of one case takes ‘

N

’ unit of time Then comparing with

N

attributes of total number of stored cases (

M

) in the CB takes “

M\times N

” unit of time Total time taken for Calculation

=

“

M\times N

” units of time

Let, Case-Base holds total ‘

M

’ number of cases and each case contains ‘

N

’ number of attributes After, repartition, let, Case-Base is divided into two Sub-Case-Bases and each of which holds ‘

P

’ number of cases So, when new query case arrives, system takes ‘

T

’ time to select Sub-Case-Base using decision tree. Here,

T=

(time taken to search colour

*

time taken to search Sub-Case-Base) If comparing with

N

attributes of one case takes ‘

N

’ unit of time Then comparing with

N

attributes of total ‘

P

’ number of stored cases (

M

) in one sub CB takes “

P\times N

” unit of time Total time taken for Calculation

=

“(

T

+

(

P\times N

))”; units of time

Example: Let,

M=

N=

5 Total time taken for Calculation

=

“

M\times N

” units of time 10

\times

=

70 units of time As, R4 model has no Repartition stage, so every time, total calculation time will be same.

Value of $M$	10
Total calculation time	50

Example: Let,

M=

N=

5 After, repartition, let, Case-Base is divided into minimum two Sub-Case-Bases and each of which holds ‘

P

’ number of cases So,

P=5

Here,

T=

(time taken to search colour

\times

time taken to search Sub-Case-Base)

=2\times 1=2

Total time taken for Calculation

=

“

(T+(P\times N))

” units of time

=2+(5\times 5)=27

units of time

Value of $M$	1	2	3	4	5	6	7	8	9
Total calculation time	7	12	17	22	27	32	37	42	47

Figure 2.

Comparison of total calculation time between R4 and R5 Model in ESECFDS.

Observation: Prior works have shown that systems concerned with car fault detection were very limited. Table 1 shows that different methods are used to diagnose different types of faults or failures of different systems. A few papers were developed or proposed for vehicle fault diagnosis. Table 2 shows the survey based on various Case-Based Reasoning Methodologies used in Automobile Domain. The survey that had been done is mainly focusing on which types of CBR models are used for implementing CBR based systems. And most of the work is based on the R4 model of CBR. Table 3 shows a comparison between R4 and R5 Model expressing that the R5 model works better than the R4 model in the CBR based System.

5. The proposed model

5.1 Architecture of ESECFDS

This paper proposes a model of Excessive-Smoke-Emitting-Car Fault Diagnosis System (ESECFDS) which is a model for question-response. The proposed ESECFDS model follows the question-response model where the user submits his query as a form of request and the ESECFDS provides the solution in form of responses. Figure 3 shows the architectural components of the ESECFDS model. It has three main modules; Administrator module, computational module, and CBR module. The Case-Base (CB) and the Interface are the foundation of the model. The Case-Base is repartitioned in Sub-Case-Bases and the user and the administrator are two elements of the interface.

User: The user can enter the ESECFDS system and posts a question directly to the system inside the User Interface.

Administrator: He is the knowledge expert and maintainer of the system. To access the entire Case-Base and cases as well, he should have a personal login ID and password. He is in charge of managing the whole system, including the Case-Base and cases as well. Cases may be added, removed, revised and Sub-Case-Bases can also be changed by the Administrator.

The administrator module assists administrators in maintaining the whole system. The administration module maintains, repartition, revises and retains corresponding process cycles of the CBR R5 model as explained in Section 2.

The computational module can select the right Sub-Case-Base using a decision tree and calculate the similarity percentage between the new case and stored cases within the selected Sub-Case-Base. The Computational Module uses the Jaccard similarity method for similarity calculation.

CBR module takes the responsibility to retrieve the most similar case with the solution associated with it from the Sub-Case-Base and displays it to the appropriate User.

Figure 3.

The architectural components of ESECFDS.

5.2 Context diagram of the ESECFDS

Figure 4 shows the context diagram related to the ESECFDS architecture, that simplifies the ESECFDS processing flow.

A user enters the ESECFDS and the system will then provide him a smoke-related question form. Next, in the user component of the Interface, the user can post a question directly to the system. The system only accepts the question form if the user fills it correctly. The system considers the query as a new query case after acceptance, which is made up of attributes-values, and the system is ready to process the query case. At the beginning of processing, the Computational Module takes the query case and identifies its attributes with values. Then, according to the smoke colour value (attribute-value) of the query case, the Computational Module selects the right Sub-Case-Base with the help of the decision tree.

The decision tree helps the Computational Module to take decisions as to which Sub-Case-Base is right for which smoke-colour value. Now, after selecting the Sub-Case-Base, the Computation Module calculates Jaccard similarity between new query case and stored cases within that Sub-Case-Base. At last, after all, calculations, the CBR Module takes the responsibility to retrieve the most similar case with the solution associated with it and displays it to the appropriate User. CBR Module also stores this user query with new solutions to the CB and updates it accordingly.

Figure 4.

Context diagram of the ESECFDS.

5.3 Use case diagram of the ESECFDS

Figure 5 demonstrates the use case diagram identifying many actions performed by ESECFDS. In Fig. 5, the actor of the use case diagram is the ESECFDS and related actions have been shown. The stepwise description is as follows.

Figure 5.

Use Case Diagram of the ESECFDS.

Check submitted query form – ESECFDS reads the Submitted Query Form from the interface and starts checking each field of the form. ESECFDS accepts the submitted form if all fields of the form are properly filled-up.

Identify query attributes-values – After acceptance of the query form, ESECFDS reads all submitted values and identifies query case’s all Attributes-Values.

Make the query set – The ESECFDS separates each Attribute-Value with a comma and makes the Query Set with it.

Select the Sub-Case-Base – The ESECFDS reads the smoke-colour value from the user’s submitted data and starts matching the value in the decision tree. According to the result of the decision tree, it indicates which Sub-Case-Base should be selected; ESECFDS selects the right Sub-Case-Base. Decision tree utilization is “include” type use case which has been defined in Fig. 5.

Apply the Jaccard similarity method – The ESECFDS applies the Jaccard Similarity Method between the prepared Query Set and all the Case Sets within the Sub-Case-Base. After all similarity calculations, ESECFDS finds the most similar case. The selection of most similar cases is “include” type use case which has been defined in Fig. 5.

Retrieve solutions – The ESECFDS uses SQL code and retrieves the solutions of that most similar case. After that, it displays those solutions to the user through the user interface. SQL code utilization and display solution are “include” type use cases which have been defined in Fig. 5.

5.4 Basic functionality of the ESECFDS

Basic functionality of the ESECFDS is shown in Fig. 6.

Figure 6.

Basic functionality of the ESECFDS.

6. Case-Base and case representation

The concept of ESECFDS has been applied to diagnose technical problems associated with cars emitting excessive smoke. The core component of this system is Case-Base (CB) and case representation in it.

Table 4
All cases, adopted from the internet

Case number	Smoke colour	Emission time	Air filter condition	Engine power	Weather	Diagnosis
Case 1	Black	Starting engine	Good	Normal	Normal	If it cleared up while engine warmed up then its ok otherwise engine sensor is faulty
Case 2	Black	Starting engine	Dirty	Normal	Normal	If it cleared up while engine warmed up then its ok otherwise engine sensor is faulty and also air filter have to be replaced
Case 3	Black	Driving	Good	Normal	Normal	Fault in sensors. Fuel injector fault, fuel pressure regulator fault
Case 4	Black	Driving	Dirty	Low	Normal	Fault in sensors. Fuel injector fault, fuel pressure regulator fault, piston rings are probably bad and air filter have to be replaced
Case 5	Black	Accelerating	Good	Normal	Normal	Engine piston rings are damaged, faulty injectors
Case 6	Black	Accelerating	Dirty	Normal	Normal	Fault in air filter, engine piston rings are damaged, faulty injectors
Case 7	White	Starting engine	Good	Normal	Cold	If it cleared up while engine warmed up then its ok otherwise head gasket problem, engine sensor problem and have to condensation steaming off
Case 8	White	Driving	Dirty	Low	Normal	Head gasket problem, replace air filter, piston rings are probably bad
Case 9	White	Driving	Dirty	Low	Cold	Head gasket problem, replace air filter, piston rings are probably bad and have to condensation steaming off
Case 10	White	Accelerating	Good	Normal	Normal	Problem in transmission fluid or burning coolant
Case 11	Blue	Accelerating	Good	Normal	Normal	The valve stem seals may be bad
Case 12	Blue	Driving	Dirty	Low	Cold	The piston rings are probably bad, air filter have to be replaced, and have to condensation steaming off
Case 13	Blue	Starting engine	Good	Normal	Normal	Excess oil consumption and the valve stem seals may be bad, damaged glow plug
Case 14	Blue	Starting engine	Dirty	Low	Normal	Excess oil consumption and the valve stem seals may be bad and air filter and piston rings are probably bad, damaged glow plug

Table 5

Sub-Case-Base1 with cases those have ‘Smoke Colour $=$ ‘Black”

Case number	Smoke colour	Emission time	Air filter condition	Engine power	Weather	Diagnosis
Case 1	Black	Starting engine	Good	Normal	Normal	If it cleared up while engine warmed up then its ok otherwise engine sensor is faulty
Case 2	Black	Starting engine	Dirty	Normal	Normal	If it cleared up while engine warmed up then its ok otherwise engine sensor is faulty and also air filter have to be replaced
Case 3	Black	Driving	Good	Normal	Normal	Fault in sensors. Fuel injector fault, fuel pressure regulator fault
Case 4	Black	Driving	Dirty	Low	Normal	Fault in sensors. Fuel injector fault, fuel pressure regulator fault, piston rings are probably bad and air filter have to be replaced
Case 5	Black	Accelerating	Good	Normal	Normal	Engine piston rings are damaged, faulty injectors
Case 6	Black	Accelerating	Dirty	Normal	Normal	Fault in air filter, engine piston rings are damaged, faulty injectors

Table 6

Sub-Case-Base2 with cases those have ‘Smoke Colour $=$ ‘White”

Case number	Smoke colour	Emission time	Air filter condition	Engine power	Weather	Diagnosis
Case 1	White	Starting engine	Good	Normal	Cold	If it cleared up while engine warmed up then its ok otherwise head gasket problem, engine sensor problem and have to condensation steaming off
Case 2	White	Driving	Dirty	Low	Normal	Head gasket problem, replace air filter, piston rings are probably bad
Case 3	White	Driving	Dirty	Low	Cold	Head gasket problem, replace air filter, piston rings are probably bad and have to condensation steaming off
Case 4	White	Accelerating	Good	Normal	Normal	Problem in transmission fluid or burning coolant

Table 7

Sub-Case-Base3 with cases those have ‘Smoke Colour $=$ ‘Blue”

Case number	Smoke colour	Emission time	Air filter condition	Engine power	Weather	Diagnosis
Case 1	Blue	Accelerating	Good	Normal	Normal	The valve stem seals may be bad
Case 2	Blue	Driving	Dirty	Low	Cold	The piston rings are probably bad, air filter have to be replaced, and have to condensation steaming off
Case 3	Blue	Starting engine	Good	Normal	Normal	Excess oil consumption and The valve stem seals may be bad, damaged glow plug
Case 4	Blue	Starting engine	Dirty	Low	Normal	Excess oil consumption and the valve stem seals may be bad and air filter and piston rings are probably bad, damaged glow plug

6.1 Case-Base of the ESECFDS

To handle the queries, the ESECFDS must have a relevant Case-Base that consists of problem descriptions (symptoms or features) related to excessive smoke problems of cars and probable solutions (diagnosis results). For car faults on smoke emission only, we adopt, different cases from different websites of the internet, referred from [28, 29, 30, 31, 32], for experimentation. Cases shall be further enhanced. Table 4 shows all cases, adopted from the internet. Each case has five attributes: Smoke colour, emission time, air filter condition, engine power, and weather. Out of these attributes, smoke colour is the most important attribute because each different smoke colour indicates different car problems. So, to diagnose the problems about excessive smoke emission from the exhaust pipe of the car, at first car users have to notice the exact smoke colour coming from the exhaust pipe.

As the ESECFDS follows the R5 Model to minimize the execution time of the fetching solution; so the system’s Case-Base is repartitioned or divided into Sub-Case-Bases. Here, according to the smoke-colour, the whole cases in Table 4 are repartitioned or divided into three different tables. Where in Table 5 (Sub-Case-Base1), cases have ‘Smoke Colour $=$ ’Black”, in Table 6 (Sub-Case-Base2), cases have ‘Smoke Colour $=$ ’White”, and in Table 7 (Sub-Case-Base3), cases have ‘Smoke Colour $=$ ’Blue”. That means the ESECFDS-Case-Base is repartitioned with three Sub-Case-Bases. As different smoke-colour indicates different car problems, so cases with different smoke colours are stored in different Sub-Case-Bases. Following Fig. 7 shows the example of ESECFDS -Case-Base. Here, the whole Case-Base is divided into 3 Sub-Case-Bases where Sub Case-Base-1 stores all the cases whose smoke colour value are “Black”, Sub-Case-Base-2 stores all the cases whose smoke colour value are “White”, and Sub-Case-Base-3 stores all the cases whose smoke colour value are “Blue”.

Figure 7.

Case-Base (CB) of ESECFDS.

6.2 Case representation in Case-Base

A Case-Based reasoning system’s efficiency depends very much on the case representation in the CB. Initially, the Administrator is responsible to populate the Case-Base with different cases and also divide the Case-Base into Sub-Case-Bases. According to that division Administrator stores a decision tree into the Computational Module that helps the system to find the exact solution from the exact Sub-Case-Base. The decision tree helps the Computational Module to decide as to which Sub-Case-Base is right for which smoke- colour value. Figure 8 shows the decision tree that ESECFDS Computational Module holds. The administrator should follow some rules to populate the Case-Base.

Different cases with different smoke colour should be stored in different Sub-Case-Bases.

The value of each Attribute should be a nonempty, single word, and correctly spelled.

Figure 8.

Decision tree in ESECFDS.

At the stage of ‘Retain’, ESECFDS will also follow the repetition rules. That means, the new cases with retrieved solutions from the Case-Base, will store in the appropriate Sub-Case-Bases. For example, the new cases with smoke colour value are Black, are to be stored in Table 5 Sub-Case-Base1; similarly, the new cases with smoke colour value are White, are to be stored into another Sub-Case-Base2 Table 6 and the new cases with smoke colour value are Blue, are to be stored into another Sub-Case-Base3 Table 7.

7. The proposed methodology

This paper proposed ESECFDS i.e. is a question- response system. A user enters the ESECFDS system and then the system will provide him a query form related to smoke problems. Next, the user can post a query directly to the system in the user component of the Interface. In this way, the system gets a new query case that is made of attributes-values, and then the system gets ready to calculate the similarity calculations between the new query case and stored cases in the Case-Base. To perform the calculations, initially, according to the smoke colour value (attribute-value) of the query case, the system finds the right Sub-Case-Base with the help of the decision tree. After that, the system uses the Jaccard similarity method to calculate the similarity. Jaccard similarity calculation method performs with a pair of cases (new query case and stored cases). After calculating the similarity between new query case and stored cases in a particular Sub-Case-Base, ESECFDS finds the highest similarity score between two sets of cases (new query case and particular one stored case) among all sets of cases. As the higher percentage value indicates a higher similarity between the two sets of cases. So, the case with the highest Jaccard similarity is considered as the most similar case. Next, the system retrieves the solutions of that stored case and displays it to the user. Now the system considers solutions are the solutions for the query case. So, the system upgrades that particular Sub-Case-Base (from where the most similar case found for the new query case) with the new query case with the adopted solution.

Jaccard Similarity Method: Calculation is done with a pair of sets in the Jaccard similarity calculation method. And it is also applied to documents as seen in [33]. The calculation focuses similarity between finite sample sets, and is formally defined as the size of the intersection divided by the size of the union of the sample sets. Let, $A$ and $B$ are two finite sets, the mathematical representation of the Jaccard Similarity is written as:

$\displaystyle J(A,B)=\frac{|A\cap B|}{|A\cup B|}=\frac{|A\cap B|}{|A|+|B|-|A% \cap B|}$ (1)

For example, let, $A$ and $B$ are two sets of words. Let,

$\displaystyle A=\text{[I, am, Sam]}$ $\displaystyle B=\text{[Ram, I, am]}$ $\displaystyle C=\text{[Sam, is, he]}$

So, according to the mathematical equation, Jaccard Similarity between $A$ and $B$ sets:

$\displaystyle(A\cup B)=\text{[I], [am], [Sam], [Ram]}$ $\displaystyle(A\cap B)=\text{[I], [am]}$

Then, Jaccard similarity $=$ $[(A\cap B)/(A\cup B)]$ $=$ 2/4 $=$ 0.5.

To get percentage $=$ 2/4 $\times$ 100 $=$ 50%.

And, according to the mathematical equation, Jaccard Similarity between $A$ and $C$ sets:

$\displaystyle(A\cup C)=\text{[I], [am], [Sam], [is], [he]}$ $\displaystyle(A\cap C)=\text{[Sam]}$

Then, Jaccard similarity $=$ $[(A\cap C)/(A\cup C)]$ $=$ 1/5 $=$ 0.2.

To get percentage $=$ 1/5 $\times$ 100 $=$ 20%.

As the greater percentage value reveals a greater similarity between two sets as referred to in [34], so here set A is more similar to set $B$ than $C$ .

7.1 Jaccard similarity method in ESECFDS

In ESECFDS, computational module holds Jaccard similarity method to calculate the similarity between sets of cases. Following steps show how computational module uses Jaccard similarity method in ESECFDS:

Step 1: Whenever user posts a new query, the Computational Module considers the new query as a new case and identifies each attributes-values of that new case and makes a set with these attributes-values of the new case.

Let, user posts a new query, the new query case holds attributes-values as:

Smoke colour $=$ black, emission time $=$ accelerating, air filter condition $=$ good, engine power $=$ low, weather $=$ cold, computational module makes a set with these attributes-values of the new case.

New query case (NQC) $=$ [black, accelerating, good, low, cold]

Step 2: According to the stored decision tree (referred to in Fig. 8), Computational Module selects a particular Sub-Case-Base. As the New Query Case’s Smoke Colour $=$ Black, So, Computational Module selects Table 5 Sub-Case-Base1 for further calculations.

Step 3: Now, computational module fetches each stored cases from the Sub-Case-Base1 and makes sets with their attributes-values. After that the module calculates similarity between new case set and stored case sets. Jaccard similarity method is applied for the similarity calculation.

For example, new query case (NQC) $=$ [black, accelerating, good, low, cold]

And from Table 5 Sub-Case-Base1, computational module makes sets as:

$\displaystyle\text{Case 1}=\text{[black, starting engine, good, normal,}\text{% normal]}$ $\displaystyle\text{Case 2}=\text{[black, starting engine, Dirty, normal,}\text% {normal]}$ $\displaystyle\text{Case 3}=\text{[black, driving, good, normal, normal]}$ $\displaystyle\text{Case 4}=\text{[black, driving, dirty, low, normal]}$ $\displaystyle\text{Case 5}=\text{[black, accelerating, good, normal,}\text{% normal]}$ $\displaystyle\text{Case 6}=\text{[black, accelerating, dirty, normal,}\text{% normal]}$

Now, Jaccard similarity method is applied for the similarity calculation between (NQC) and all sets of cases.

Step 3.1: Computational Module fetches Case 1, calculate similarity calculation between (NQC) and Case 1.

Jaccard similarity Percentage of (Case 1, NQC) $\displaystyle=[(\text{Case 1}\cap\text{NQC})/(\text{Case}1\cup\text{NQC})]% \times 100$ $\displaystyle=(2/8)\times 100=25{\%}$

Step 3.2: Computational Module fetches Case 2, calculate similarity calculation between (NQC) and Case 2.

Jaccard similarity Percentage of (Case 2, NQC) $\displaystyle=[(\text{Case 2}\cap\text{NQC})/(\text{Case 2}\cup\text{NQC})]% \times 100$ $\displaystyle=(1/8)\times 100=12.5{\%}$

Step 3.3: Computational Module fetches Case 3, calculate similarity calculation between (NQC) and Case 3.

Jaccard similarity Percentage of (Case 2, NQC) $\displaystyle=[(\text{Case 3}\cap\text{NQC})/(\text{Case 3}\cup\text{NQC})]% \times 100$ $\displaystyle=(2/7)\times 100=28.57{\%}$

Step 3.4: Computational Module fetches Case 4, calculate similarity calculation between (NQC) and Case 4.

Jaccard similarity Percentage of (Case 2, NQC) $\displaystyle=[(\text{Case 4}\cap\text{NQC})/(\text{Case 4}\cup\text{NQC})]% \times 100$ $\displaystyle=(2/8)\times 100=25{\%}$

Step 3.5: Computational Module fetches Case 5, calculate similarity calculation between (NQC) and Case 5.

Jaccard similarity Percentage of (Case 2, NQC) $\displaystyle=[(\text{Case 5}\cap\text{NQC})/(\text{Case 5}\cup\text{NQC})]% \times 100$ $\displaystyle=(3/6)\times 100=50{\%}$

Step 3.6: Computational Module fetches Case 6, calculate similarity calculation between (NQC) and Case 6.

Jaccard similarity Percentage of (Case 2, NQC) $\displaystyle=[(\text{Case 6}\cap\text{NQC})/(\text{Case 6}\cup\text{NQC})]% \times 100$ $\displaystyle=(2/7)\times 100=28.57{\%}$

In this way, Jaccard similarity Percentages are calculated for all stored cases in the Sub-Case-Base1.

Step 4: Now, CBR module finds the stored case with the highest Jaccard similarity percentage with the NQC. Next, the module retrieves the solutions of that stored case and displays it to the user. Now the system considers solutions are the solutions for the query case.

For example, Case 5 has the highest Jaccard similarity percentage with the NQC. So, the CBR module retrieves the solutions of Case 5 and considers that solutions are the solutions for the NQC.

So, for the new query case (NQC), Smoke colour $=$ black, emission time $=$ accelerating, air filter condition $=$ good, engine power $=$ low, weather $=$ cold, solution is: Fault in plugs, fuel filter.

Step 5: CBR module stores this user query with new solutions to the CB and updates it accordingly.

For example, Table 5 Sub-Case-Base1 is updated with,

Case number	Smoke colour	Emission time	Air filter condition	Engine power	Weather	Diagnosis
Case 7	Black	Accelerating	Good	Low	Cold	Fault in plugs, fuel filter

8. The proposed algorithm with simulation results

Step 1: Start

Step 2: Read Input Case NQC.

Step 3: Identify each Input Attribute-value

of NCQ

Step 4: Initialize each Input Attributes-

values of NCQ

Smoke Colour Value as ’S’

Emission Time value as ’Em’

Air Filter Condition value as ’A’

Engine Power value as ’En’ and

Weather value as ’W’

Step 5: Declare NQC as String Array as NQC[].

Step 6: Declare values in array

NQC[] = {S, Em, A, En, W}

Step 7: ’S’ is checked from NQC, and select

the particular Sub-Case-Base using

Decision tree

Step 8: If ’S’ = "Black"

\\ Select Sub-Case-Base 1.

Let, N is the total number of cases

in Sub-Case-Base 1.

Step 8.1: Declare an Integer Array as

Result[].

Step 8.2: For i = 1 to N

Result[i]=Jaccard Similarity

Percentage between

(NQC[], Casei[])

End For

Step 8.3: large = Result[1]

For i = 1 to N

If (Result[i] >= large):

large = Result[i]

Similar_case_no = i

End If

End For

print("Highest similarity

score:", large)

print("most similar Case:",

Similar_case_no)

Step 9: Else If ’S’ = "White"

\\ Select Sub-Case-Base 2.

Let, M is the total number of cases

in Sub-Case-Base 2.

Step 9.1: Declare an Integer Array as

Result[].

Step 9.2: For i = 1 to M

Result[i]=Jaccard Similarity

Percentage between

(NQC[], Casei[])

End For

Step 9.3: large = Result[1]

For i = 1 to M

If (Result[i] >= large):

large = Result[i]

Similar_case_no = i

End If

End For

print("Highest similarity

score:", large)

print("most similar Case:",

Similar_case_no)

Step 10: Else If ’S’ = "Blue"

\\ Select Sub-Case-Base 3.

Let, P is the total number of

cases in Sub-Case-Base 3.

Step 10.1: Declare an Integer Array as

Result[].

Step 10.2: For i = 1 to P

Result[i]=Jaccard Similarity

Percentage between

(NQC[], Casei[])

End For

Step 10.3: large = Result[1]

For i=1 to P

If (Result[i] >= large):

large = Result[i]

Similar_case_no = i

End If

End For

print("Highest similarity

score:", large)

print("most similar Case:",

Similar_case_no)

End If

Step 11: Solutions of the Most Similar

case are selected as Output.

Step 12: Stop

Example 1 with simulation results:

Let, following Fig. 9 is the user query, so the system follows the above-mentioned algorithm.

Figure 9.

User query to the system.

Read input case NQC.

Identify each input attribute-value of NCQ.

Initialize each input attributes-values of NCQ smoke colour value as ‘black’.

–

Emission time value as ‘driving’.

–

Air filter condition as ‘good’.

–

Engine power as ‘low’ and

–

Weather as ‘temperate’.

Declare NQC as String Array as NQC[].

Declare values in array.

NQC[] $=$ {black, driving, good, low, temperate}.

Input Smoke Colour Value is ‘Black’ is checked from NQC, and select the particular Sub-Case-Base using Decision tree.

From the above algorithm, select Sub-Case-Base 1. Let, $N=$ 6, is the total number of cases in Sub-Case-Base 1.

Declare an Integer Array as Result[].

For $i=$ 1 to 6

Result $[i]=$ Jaccard Similarity Percentage

between (NQC[], Casei[])

End For

For $i=1$ to 5

For $j=(i+1)$ to 6

If Result $i>$ Result $j$

Most Similar Case $=$ Case $i$

Else

Most Similar Case $=$ Case $j$

End If

End For

10.

Solutions of the most similar case are selected as output.

Following Fig. 10 shows the feedback result of the system.

Figure 10.

Feedback result of system to the user.

Example 2 with simulation result:

Let, smoke colour value as ‘white’

•

Emission time value as ‘starting engine’.

•

Air filter condition as ‘good’.

•

Engine power as ‘low’ and

•

Weather as ‘cold’.

9. Performance analysis of proposed methodology

Figure 11.

Feedback result of system to the user.

The proposed methodology helps the ESECFDS to retrieve solutions to the excessive smoke emission problems of cars from the Sub-Case-Bases in minimum time. The proposed methodology uses the Jaccard Similarity Method for similarity calculation between new cases and old cases.

9.1 Performance of jaccard similarity method

As we can see from the previous related works that different Case-Based Reasoning-based papers used different methods to retrieve cases. As mentioned to in [6], nearest neighbor method and feature matching are the retrieval methods, the paper [14] suggest combining Hamming distance and Euclidean distance is the better for retrieval, in [16] decision tree learned to distinguish defective cases, the paper [21] applies nearest neighbors as the case-retrieval algorithm, and the CBR-based expert system in [23], uses K-nearest neighbor (KNN) algorithm as the case retrieval method. So, from the previous related works, it is clear that the K-nearest neighbor (KNN) algorithm is very popular as a case retrieval method in Case-Base reasoning. And KNN uses Euclidean distance method to find the distance (similarity) between new case and cases in the Case-Base.

KNN uses following simple Steps for similarity calculation in Case-Based system:

Step 1: Step 1:
Get the new case.
Step 2:
Initialize nearest neighbor’s value ‘ $k$ ’.
Step 3:
Use following Euclidean distance formula to find the distance (similarity) between new case and cases in the Case-Base.
Step 4:
Find most similar ‘ $k$ ’ cases.
Step 5:
Get the most similar case out of ‘ $k$ ’ cases.

There is a disadvantage of KNN in the CBR system. As referred to in [35], the disadvantage of the KNN is that the retrieval time increases directly with the increasing size of the Case-Base. Euclidean distance method used by KNN takes more time to calculate the similarity between two sets than Jaccard distance calculation. A comparison between the Euclidean distance method and the Jaccard Similarity Method is shown in Table 8.

Table 8
A comparison between the Euclidean distance method and the Jaccard Similarity method

Euclidean distance method Jaccard similarity method

Euclidean distance method uses following formula to find the distance (similarity) between new case and cases in the Case-Base. $d(p,q)=\sqrt{(p_{1}-q_{1})^{2}+(p_{2}-q_{2})^{2}}$ Jaccard distance method uses following formula to find the distance (similarity) between new case and cases in the Case-Base. $J(A,B)={\displaystyle\frac{|A\cap B|}{|A\cup B|}}={\displaystyle\frac{|A\cap B% |}{|A|+|B|-|A\cap B|}}$

Experiment: Calculate execution time to find Euclidian distance between two sets: Example 1: [0, 1, 2, 5, 6], [0, 2, 3, 5, 7, 9] Similarity: 17.3% execution time: 0.0006247 Sec. Example 2: [0, 1, 3], [0, 1, 2] Similarity: 100% execution time: 0.0007705 Sec. In every example Euclidian distance calculation takes more time than Jaccard distance calculation. Experiment: Calculate execution time to find Jaccard distance between two same sets: Example 1: [0, 1, 2, 5, 6], [0, 2, 3, 5, 7, 9] Similarity: 37.5% execution time: 0.0003649 Sec. Example 2: [0, 1, 3], [0, 1, 2] Similarity: 50% execution time: 0.0005041 Sec. In every example Jaccard distance takes less time than calculation Euclidian distance calculation.

As, the Case-Base of ESECFDS contains words as the attribute-values. Here, euclidean distance is not so useful and popular for text processing. Jaccard or Cosine similarities are very useful and popular for text processing.

Table 9
A comparison between the cosine similarity method and the Jaccard similarity method

Cosine similarity method Jaccard similarity method

Cosine Similarity calculates similarity be measuring the cosine of angle between two vectors. The similarity between the two strings is the cosine of the angle between these two vectors representation, and is computed as $V1\times V2/(|V1|*|V2|)$ Jaccard similarity is defined as size of intersection divided by union of two sets. The Jaccard index is then computed as $|V1$ inter $V2|/|V1$ union $V2|$ .

Cosine similarity takes total length of the vectors. This means when cosine similarity is calculated between two Sentences, if any ‘word’ repeats in the first Sentence for several times, cosine similarity changes and drops the similarity score. Jaccard similarity takes only unique set of words for each sentence / document. This means when Jaccard similarity is calculated between two Sentences, if any ‘word’ repeats in the first Sentence for several times, Jaccard similarity is not changed and the similarity score remain same.

When cosine similarity is used in the proposed methodology of ESE NewCase $=$ [’Black’, ’Accelerating’, ’good’, ’low’, ’cold’] Case 1 $=$ [’Black’, ’Starting engine’, ’good’, ’normal’, ’normal’] Cosine Similarity $=$ 34% Execution time: 0.0011395 sec NewCase $=$ [’Black’, ’Accelerating’, ’good’, ’low’, ’cold’] Case 2 $=$ [’Black’, ’Starting engine’, ’normal’, ’normal’, ’normal’] Jaccard Similarity $=$ 13% Execution Time: 0.0011074 Sec. Cosine similarity calculation takes little extra time than Jaccard similarity calculation. Similarity Calculation of cosine similarity is almost similar to Jaccard similarity When Jaccard similarity is used in the proposed methodology of ESE NewCase $=$ [’Black’, ’Accelerating’, ’good’, ’low’, ’cold’] Case 1 $=$ [’Black’, ’Starting engine’, ’good’, ’normal’, ’normal’] Jaccard Similarity $=$ 29% Execution Time: 0.0005731 sec NewCase $=$ [’Black’, ’Accelerating’, ’good’, ’low’, ’cold’] Case 2 $=$ [’Black’, ’Starting engine’, ’normal’, ’normal’, ’normal’] Jaccard Similarity $=$ 14% Execution Time: 0.0004185 Sec. Jaccard similarity calculation takes less time than cosine similarity calculation. Similarity Calculation of Jaccard similarity is almost similar to cosine similarity

The computational cost of cosine similarity to compute the similarity between two strings of length $m$ and $n$ respectively, is $O(m+n)$ The computational cost of Jaccard similarity to compute the similarity between two strings of length $m$ and $n$ respectively, is $O(m+n)$

So, from Table 8, it is clear that the Jaccard Similarity Method works in less time than the Euclidean distance method; and the Case-Base of ESECFDS contains words as the attribute-values. So, this paper prefers Jaccard Similarity Method for text-similarity calculation.
9.2 Performance of proposed methodology with text-similarity calculation methods

Euclidean distance method	Jaccard similarity method
Euclidean distance method uses following formula to find the distance (similarity) between new case and cases in the Case-Base. $d(p,q)=\sqrt{(p_{1}-q_{1})^{2}+(p_{2}-q_{2})^{2}}$	Jaccard distance method uses following formula to find the distance (similarity) between new case and cases in the Case-Base. $J(A,B)={\displaystyle\frac{\|A\cap B\|}{\|A\cup B\|}}={\displaystyle\frac{\|A\cap B% \|}{\|A\|+\|B\|-\|A\cap B\|}}$
Experiment: Calculate execution time to find Euclidian distance between two sets: Example 1: [0, 1, 2, 5, 6], [0, 2, 3, 5, 7, 9] Similarity: 17.3% execution time: 0.0006247 Sec. Example 2: [0, 1, 3], [0, 1, 2] Similarity: 100% execution time: 0.0007705 Sec. In every example Euclidian distance calculation takes more time than Jaccard distance calculation.	Experiment: Calculate execution time to find Jaccard distance between two same sets: Example 1: [0, 1, 2, 5, 6], [0, 2, 3, 5, 7, 9] Similarity: 37.5% execution time: 0.0003649 Sec. Example 2: [0, 1, 3], [0, 1, 2] Similarity: 50% execution time: 0.0005041 Sec. In every example Jaccard distance takes less time than calculation Euclidian distance calculation.
As, the Case-Base of ESECFDS contains words as the attribute-values. Here, euclidean distance is not so useful and popular for text processing.	Jaccard or Cosine similarities are very useful and popular for text processing.

Cosine similarity method	Jaccard similarity method
Cosine Similarity calculates similarity be measuring the cosine of angle between two vectors. The similarity between the two strings is the cosine of the angle between these two vectors representation, and is computed as $V1\times V2/(\|V1\|*\|V2\|)$	Jaccard similarity is defined as size of intersection divided by union of two sets. The Jaccard index is then computed as $\|V1$ inter $V2\|/\|V1$ union $V2\|$ .
Cosine similarity takes total length of the vectors. This means when cosine similarity is calculated between two Sentences, if any ‘word’ repeats in the first Sentence for several times, cosine similarity changes and drops the similarity score.	Jaccard similarity takes only unique set of words for each sentence / document. This means when Jaccard similarity is calculated between two Sentences, if any ‘word’ repeats in the first Sentence for several times, Jaccard similarity is not changed and the similarity score remain same.
When cosine similarity is used in the proposed methodology of ESE NewCase $=$ [’Black’, ’Accelerating’, ’good’, ’low’, ’cold’] Case 1 $=$ [’Black’, ’Starting engine’, ’good’, ’normal’, ’normal’] Cosine Similarity $=$ 34% Execution time: 0.0011395 sec NewCase $=$ [’Black’, ’Accelerating’, ’good’, ’low’, ’cold’] Case 2 $=$ [’Black’, ’Starting engine’, ’normal’, ’normal’, ’normal’] Jaccard Similarity $=$ 13% Execution Time: 0.0011074 Sec. Cosine similarity calculation takes little extra time than Jaccard similarity calculation. Similarity Calculation of cosine similarity is almost similar to Jaccard similarity	When Jaccard similarity is used in the proposed methodology of ESE NewCase $=$ [’Black’, ’Accelerating’, ’good’, ’low’, ’cold’] Case 1 $=$ [’Black’, ’Starting engine’, ’good’, ’normal’, ’normal’] Jaccard Similarity $=$ 29% Execution Time: 0.0005731 sec NewCase $=$ [’Black’, ’Accelerating’, ’good’, ’low’, ’cold’] Case 2 $=$ [’Black’, ’Starting engine’, ’normal’, ’normal’, ’normal’] Jaccard Similarity $=$ 14% Execution Time: 0.0004185 Sec. Jaccard similarity calculation takes less time than cosine similarity calculation. Similarity Calculation of Jaccard similarity is almost similar to cosine similarity
The computational cost of cosine similarity to compute the similarity between two strings of length $m$ and $n$ respectively, is $O(m+n)$	The computational cost of Jaccard similarity to compute the similarity between two strings of length $m$ and $n$ respectively, is $O(m+n)$

From the previous work, referred to in [36, 37], the Cosine Similarity method is very popular for the text-similarity Calculation. A comparison between the Cosine Similarity method and Jaccard Similarity Method is shown in Table 9. The table shows that the Similarity Calculation of Jaccard similarity is almost similar to cosine similarity but cosine similarity calculation takes little extra time than Jaccard similarity calculation. The worst-case time complexity is calculated to find the most similar case using the Jaccard similarity method in ESECFDS. To compare the time complexity, the Cosine Similarity method is used to find the most similar case from the Case-Base in ESECFDS and the worst-case time complexity is calculated. Figure 12 shows the worst-case time complexity comparison graph, where it is clear that the Cosine Similarity method in ESECFDS has relatively higher time complexity. That is why the Jaccard similarity method is chosen for similarity calculation in the Proposed Methodology in ESECFDS.

Figure 12.

Worst case time complexity comparison between cosine similarity method and jaccard similarity method in ESECFDS.

9.3 Performance of proposed algorithm with jaccard similarity method

The Proposed Algorithm performs efficiently with the Jaccard Similarity Method. The time complexity of the Proposed Algorithm in the Best case and Average Case is Constant: $O(1)$ , in the Worst case, is Linear: $O(N)$ , where $N$ is the input size. The following Fig. 13 shows the Time Complexity Graph of The Proposed Algorithm.

Figure 13.

The time complexity graph of the proposed algorithm in ESECFDS.

10. Limitations

This paper proposed ESECFDS i.e. is a query- response system. But there are some limitations of ESECFDS listed below:

iii) i)
As the Case-Base is divided into Sub-Case-Bases, it may take extra Space for storage.
ii)
Case-Base contains only some cases.
iii)
A case contains only five Attributes so diagnosis is based on these few attributes.

Despite the limitations, each architectural module of the ESECFDS performs efficiently. To overcome the limitations, more cases should be added to the Case-Base with the help of the Computational Module and more attributes with corresponding values should be added in the cases. Future work lies in recovering present limitations and also will upgrade ESECFDS in the diagnostic system for car faults (DSCF) that will deal with all types of car faults.
11. Conclusion and future work

The CBR Technique is an AI method that can be successfully used in the automobile industry to solve service-related technical problems. This paper proposes a CBR based Excessive-Smoke-Emitting-Car Fault Diagnostic System (ESECFDS) that solves the problems of car, emitting excessive smoke from the exhaust pipe. Users can post a question to the system about their problem. The system will return to them satisfactory feedback. This is the system where a user can get an idea to repair their car faults. The paper has discussed the architecture, implementation, and background theory of ESECFDS. Implementing the application may be effective in reducing many complexities of repairing Excessive-Smoke-Emitting car faults. The CBR cycle R4 Model of Aamodt and Plaza has been modified to R5 Model to improve the time complexity of the proposed algorithm of the system over the popular R4 Model. The proposed algorithm uses the Jaccard similarity method for similarity calculation between the new case and stored cases in the Case-Base. The time complexity of the Proposed Algorithm in the Best case and Average Case is Constant: $O(1)$ , in the Worst case, is Linear: $O(N)$ , where $N$ is the input size.

ESECFDS has some limitations; inspite of that, each architectural module performs efficiently. To overcome the limitations, more cases should be added to the Case-Base with the help of the Computational Module and more attributes with corresponding values should be added in the cases. Future work lies in recovering current limitations and will also update ESECFDS to Diagnostic System for Car Faults (DSCF), to deal with all forms of car faults.

Footnotes

Acknowledgments

The authors would like to thank the Department of Computer Science and Engineering in NIT, Durgapur (WB, INDIA), for academically supporting this research work.

References

Effects of the car on societies. Available: https://en.m.wikipedia.org/wiki/Effects_of_the_car_on_societies, accessed 2020.

Bergmann

Althoff

K.D.

Minor

Reichle

and Bach

, Case-Base reasoning introduction and recent developments, German Journal on Artificial Intelligence 23(1) (2009), 5–11.

Baghel

and Sharma

, A quick tour to understand and improve CBR performance, International Journal of Application or Innovation in Engineering & Management (IJAIEM) 3(10) (2014), 335–341.

Chen

and Min

, Research and implementation of vehicle fault diagnosis expert system, Advances in Information Sciences and Service Sciences (AISS) 5(7) (2013), 28–36.

Mostafa

S.A.

Ahmad

Md.S.

Mohammed

M.A.

and Obaid

O.I.

, Implementing an expert diagnostic assistance system for car failure and malfunction, International Journal of Computer Science Issues 9(2) (2012), 1–7.

Wen

Crossman

Cardillo

and Murphey

Y.L.

, Case-Base reasoning in vehicle fault diagnosis, International Joint Conference on Neural Networks 4 (2003), 2679–2684.

Sharma

Bhardwaj

and Shrivastava

, Car failure fuzzy fault diagnostic system based on fault tree analysis, International Journal of Theoretical & Applied Sciences 2(1) (2010), 18–21.

Murphey

Y.L.

Chen

Nasr

M.A.

Baker

Feldkamp

and Kolmanovsky

, Ensembles of neural networks with generalization capabilities for vehicle fault diagnostics, International Joint Conference on Neural Networks, (2009), 2188–2194.

Xiaolei

and Xiaobing

, Application of rough set theory in vehicle transmission system fault diagnosis, IEEE International Conference on Vehicle Electronics 1 (1999), 240–242.

10.

Wang

C.S.

, Research on extracting vehicle fault symptoms based on rough set theory in mechanical engineering, Advanced Materials Research 644 (2013), 110–114.

11.

Ming

and Taiyong

, Fault diagnosis of power transformer based on association rules gained by rough set, The 2nd International Conference on Computer and Automation Engineering (ICCAE), (2010), 123–126.

12.

Agrawal

A.J.

and Kakde

O.G.

, Semantic analysis of natural language queries using domain ontology for information access from database, I J Intelligent Systems and Applications 12 81–90, 2013.

13.

Jindal

Aggarwal

Jain

and Verma

, Approach towards car failure diagnosis-an expert system, International Journal of Computer Applications 1(23) (2010), 60–62.

14.

Song

Liu

and Zhu

, Research of CBR retrieval method based on rough set theory, 6th IEEE International Conference on Software Engineering and Service Science (ICSESS), (2015), 990–993.

15.

Xiong

Funk

and Olsson

, Case-Based reasoning supports fault diagnosis using sensor information, International Journal of COMADEM 16(4) (2013), 63–68.

16.

Selvamani

B.R.

and Khemani

, Decision tree induction with CBR, International Conference on Pattern Recognition and Machine Intelligence 3776 (2005), 786–791.

17.

Kher

Chande

P.K.

and Sharma

P.C.

, Automobile engine fault diagnosis using neural network, Conference: Intelligent Transportation Systems, (2001), 492–495.

18.

Samantaray

S.R.

, l. Kamwa, and Joos

, Decision tree based fault detection and classification in distance relaying, International Journal of Engineering Intelligent Systems for Electrical Engineering and Communications 18(2) (2011), 1–9.

19.

Shoukun

Xin

Xuan

and Ma

, Research of Case-Based reasoning method based on rough sets, IEEE International Conference on Intelligent Computing and Intelligent Systems, (2009), 416–419.

20.

Houeland

T.G.

, An Efficient Random Decision Tree algorithm for Case-Based reasoning systems, International Florida Artificial Intelligence Research Society Conference, (2011), 401–406.

21.

Tsai

Y.T.

, Applying a Case-Based reasoning method for fault diagnosis during maintenance, Journal of Mechanical Engineering Science 223(10) (2009), 2431–2441.

22.

Ahmed

, Computer fault diagnosis system using Case-Based reasoning approach, International Journal of Latest Research in Science and Technology 3(2) (2014), 1–6.

23.

Chattopadhyay

Banerjee

Rabhi

F.A.

and Acharya

U.R.

, A Case-Based reasoning system for complex medical diagnosis, The Journal of Knowledge Engineering 30(1) (2013), 12–20.

24.

Zhang

Zhou

and Liu

, Research of automobile quality fault diagnosis method based on CBR, Applied Mechanics and Materials 130-134, (2012), 313–316.

25.

Diana Wong

M.L.

, Integrated reasoning approach for car faulty diagnosis, International Journal of Computer and Information Engineering 5(3) (2011), 232–237, .

26.

Vacek

Gindele

Zöllner

J.M.

and Dillmann

, Situation classification for cognitive automobiles using Case-Based reasoning, IEEE Intelligent Vehicles Symposium, (2007), 704–709.

27.

Vacek

and Gindele

Zöllner

J.M.

and Dillmann

, Using Case-Based reasoning for autonomous vehicle guidance, IEEE/RSJ International Conference on Intelligent Robots and Systems, (2007), 4271–4276.

28.

Common Causes for Smoking Vehicles or Visible Exhaust. Available: http://www.nctcog.org/trans/air/smoking/causes.asp, accessed 2009.

29.

DriverSide. Available: https://www.driverside.com/car-diagnostic/hyundai-santa_fe_sport-2013-125709-304981-0?style_id=821677&mgs=1, 2, accessed 2020.

30.

The motor company. Available: https://motorcompanyleicester.co.uk/what-causes-black-smoke-from-diesel-engine/#:∼:text=Black%20smoke%20indicates%20that%20the,mixture%20of%20fuel%20and%20air.&text=This%20faulty%20process%20forms%20solid,Dirty%20EGR%20valve. accessed 2020.

31.

CarBibles. Available: https://www.carbibles.com/white-smoke-from-exhaust/. accessed 2018.

32.

Car From Japan. Available: https://carfromjapan.com/article/car-maintenance/blue-smoke-from-exhaust/. accessed 2019.

33.

Instructor: Jeff Phillips

, University of Utah, Jaccard Similarity and Shingling, CS 6955 Data Mining; Spring 2013

34.

lStatistics How

To.

,Available: http://www, statisticshowto.com/ Jaccard-index/. accessed 2018.

35.

Cunningham

, CBR: Strengths and weaknesses, 11th International Conference on Industrial and Engineering Applications of Artificial Intelligence and Expert Systems 2 (1998), 517–524.

36.

Dejun

Zhen

Jianguo

and Xianping

, Phased array antenna fault diagnosis based on subarray testing and cosine similarity, 2015 IEEE 12th International Conference on Electronic Measurement & Instruments, (2015), 202–206.

37.

Hao

and Gang

, A novel diagnosis method for intelligent IETM platform based on cosine similarity and fuzzy semantic inference, 2016 Prognostics and System Health Management Conference (PHM-Chengdu), (2016), 1–6.

Development of Excessive-Smoke-Emitting-Car Fault Diagnostic System (ESECFDS) based on Case-Base Reasoning (CBR) Methodology

Abstract

Keywords

1. Introduction

2. Case-Base Reasoning Methodology

4. Comparative analysis

4.1 Comparative analysis of previous works with their outcomes

Table 1 Comparative study of previous related works with their outcomes

Table 2 Comparative previous work analysis of various Case-Based Reasoning Methodologies used in automobile domain

Table 3 Comparison analysis of CBR R4 model and R5 model in ESECFDS

5.1 Architecture of ESECFDS

Table 4 All cases, adopted from the internet

8. The proposed algorithm with simulation results

Footnotes

Acknowledgments

References

Table 1
Comparative study of previous related works with their outcomes

Table 2
Comparative previous work analysis of various Case-Based Reasoning Methodologies used in automobile domain

Table 3
Comparison analysis of CBR R4 model and R5 model in ESECFDS

Table 4
All cases, adopted from the internet