Microprocessors and Microsystems 80 (2021) 103538

Contents lists available at ScienceDirect

Microprocessors and Microsystems

journal homepage: www.elsevier.com/locate/micpro

Software Reliability Growth Fault Correction Model Based on Machine

Learning and Neural Network Algorithm
Liya Li a, b
a
School of Mathematics and Statistics, Central China Normal University, Wuhan, Hubei, 430079, China
b
School of Mathematics and Economics, Hubei University of Education, Wuhan, Hubei, 430061, China

A R T I C L E I N F O A B S T R A C T

Keywords: Dependence is one of programming quality points can be quantified. Software Reliability Growth Model (SRGM)
Software Reliability used to study the reliability in various difficult times available. In all cases the test execution over time, the
Fault Correction traditional time-sensitive SRGM may not be accurate. To overcome this problem, SRGM practice rather than
Machine Learning
using the test of time. Previously, the proposed limited testing athletic ability, but with unlimited test time, test
Neural Network Algorithm
motion becomes infinite and may not make sense. Then, this method presents endless test execution work related
to older models of Neural Heterogeneous Poisson Process (NHPP) of. Programming model disappointment with
the proposed information to prepare Artificial Neural Network (ANN). They can imagine a lot of the load placed
on a similar model to represent the previous disappointment information. Use AI model selection method can
adequately represent the past and future load fitting location information. Consider using common-sense in­
formation programming disappointed indicators to show the proposed number of Testing Exertion Flow (TEF)
and SRGM also successfully depicted disappointment extensive information, Machine Learning (ML) and ANN
improved the accuracy of the conventional boundary estimation as compared with the discharge time and can be
used to ensure that programming.

1. Introduction strategies to determine the best AI of the proposed load model, the
model is also suitable for past and future information. Based on the well-
The development programming framework since many of these known SRGM testing, investigation and analysis of the proposed model
SRGM has been proposed as the model parameters and non-parametric of integrity Based. In all cases, always use ANN boundary assessment,
model. SRGM reason traditional model-dependent ability to follow an because it increases the accuracy of the assessment of the border, and
exponential average of the S-shaped development or development sys­ has a better fit than the traditional model of measurable parameters.
tem. SRGM is presented in the TEF, as identified and adjusted the With zero deformation and popular development, reliability prediction
problem depends on the work done by such experimental work used, the is becoming increasingly important programming projects. SRGM there
time spent on the PC as well as the consumption of various assets. I’m. are a lot of In any case, work under different conditions. Recently,
Test time or scheduled time. Based on the previously proposed the use of artificial neural networks have been used for programming unwavering
an exponential function SRGM tired, Rayleigh, or propagation calcula­ quality assessment and the development of reliability predictions. For
tion to determine the work fatigue test (TEF), to indicate that utilization the most current written tests, each troubleshooting tasks are deemed
during fatigue testing. Absolutely desirable to perform tests for an in­ necessary comparative tests. However, given its multi-faceted, may
definite period of time is a paradox. Two stage show the Fig. 1 Non- require different test work to identify and eliminate the disadvantages of
Homogeneous Poisson Process (NHPP) proposed unwavering quality different types.
development model, which relies on the distribution of powers of, the The multifaceted nature of the defect, which is divided into basic,
model is a graphical model allows the information chart before fitting difficult and complex categories. In this method, we apply the technique
boundary assessment. to generate a nerve tissue takes into account the different levels of
In this article, we recommend the use of TEF and SRGM the number complexity of programming a defect reliability development model.
of distribution power to represent and provide unlimited TEF is SRGM. Calculated on behalf of test team mastered the skills of learning will be
Boundary assessed using Artificial Neural Network (ANN), and apply used to demonstrate the proposed model. The proposed model is for

E-mail address: [email protected].

https://doi.org/10.1016/j.micpro.2020.103538
Received 30 October 2020; Received in revised form 19 November 2020; Accepted 23 November 2020
Available online 27 November 2020
0141-9331/© 2020 Elsevier B.V. All rights reserved.
L. Li Microprocessors and Microsystems 80 (2021) 103538

simple problems, develop evacuation measures are unified, and for the programming to calculate the cost of defect prediction using subtle
difficult and complex disadvantage, because as the conduct of exami­ boost. To enhance the accuracy of calculation of the cost subtle than
nations, the development exile group learning, thus providing a measure acceptable various calculations. [18] Pointed out, looking for the most
exile. We are expected to follow the trends calculated. The proposed modules expected to perform are correct, but some module is defective.
model is another NHPP model has been applied to real business They adopted a program to find the number of defects in a particular
improvement program cited two disappointing / defect by the exclusion module.
of information collection and approved, evaluation and comparison. The It has been proposed Software Reliability Growth Model (SRGM) as a
results show that the model has the computing power of the proposed continuous-time model for planning and running tests [19], and the
program to improve the applicability of disappointment / evacuation need to develop discrete models to fit the discrete data. They used the
information shortcomings. class-level metrics. Analysis is one of the key requirements analysis.
They need to receive from the various stakeholders, there needs [20],
2. Related works and to analyze their relevance to the final product.

The product is a measure of the model and the fact that, for identi­ 3. Proposed methodology
fying the product [1] frame number of defects, most of the desired size
models and unpredictable utilization metrics to detect defects. They are The methods are classified by the attached method. This section
discovering, presents a large and complex multi-fact model, in order to presents the proposed testing. Area software reliability growth model
discover isolated and unpredictable metric. Disorders the effectively put forward recommendations. This method is described in Fig. 1, to
open model found by using the size and disadvantages [2] measure verify the validity of the proposed model. Use ANN regional assessment
unpredictable. Description product defect prediction strategy is to find boundary. For selecting an appropriate model for the proposed load AI
defects to improve product quality and outstanding reliability of the programs.
method [3]. They use the methods of measurement method level. Shows the Fig. 2 survey demonstrates the ANN and described in the
The idea nerve tissue is being used strictly. Neural tissue provides an MLT region 1, a proposed use of a model, in particular to ensure that the
important method of [4], referred to as a Radiation Basis Function (RBF) discharge time programming. Executive summary and end.
[5]. RBF main function is to find the problem and make the product
more accurately. Quantity [6] Object-oriented programming framework
for predicting product defects. Polymorphism and traditional 3.1. Software reliability growth model (SRGM) - log-power testing effort
object-oriented [7] highlights an important frame. function (LPTEF)
Utilizes two key tissues [8] – Multi-Layer Perceptron (MLP) is used to
manage and coordinate the fault module defect, because various types of The Method do not recommend extended with new shiny SRGM, but
defects have demonstrated the ability of radiation based network. Used the proposal extends the excellent work done by analysts in the past, and
in 8. Xing et al. Fig. 1 shows the importance of the Support Vector using the work of several power test run for the model timing. In the old
Machine (SVM) model [9,10]. When only get very little information, (G-O) Goel-Okumoto SRGM, the independent variable time "t" to run in
method will use this SVM model. The SVM is an important sequence place of a power of test "(t)" working time through the application
information method. SVM priority to the accuracy of various methods, associated with NHPP model. Make changes as NHPP model. If (t) is the
[11] identifying a defective module. Measurement level typically used average energy investment capacity of logarithmic power testing t can
for programming the expected defects. The authors used three [12] joint now change the Goel-Okumoto model in the following manner (t):
ventures. As a result of examination of these various activities, pre­ μ(t) = (a(1 − )(( − b ∗ W(t)))) (1)
sumably, the necessity for measuring an important predictive pro­
gramming defects [13] function. W(t) = n ∗ (ln)c(1 + t) (2)
In another article, the authors describe SVM ability to find product
variants, and presentation with the characteristic [15] models were If (t) during all were tested exhausted energies (0, x], the a normal
compared. Model they created with the help of SVM has higher accuracy number of perceived programming failed, and, c and n are constants.
than other models [14]. For four NASA datasets, they tend to confirm the Mean Capacity (t) having Log Power TEF of SRGM is: method do not
ability of SVM can predict incomplete programming module, and can recommend the use of spices for this purpose and span SRGM, and the
differentiate the product model and the fact that eight AI model [16]
defect prediction representation.
This information comes from [17] Promise warehouse with unlim­
ited access to public data sets. Experts use a variety of programming
methods (such as Lines of Code (LOC)) associated with the time and
money to arrange to build a higher quality of programming.
Programming flaws expectations model module fall into two cate­
gories. The module is defective and intact. They discussed a variety of

Fig. 2. proposed architecture of software reliability growth fault correction

Fig. 1. Two level stage SRGM filtering system. model based MLT and ANN.

2
L. Li Microprocessors and Microsystems 80 (2021) 103538

proposed expansion of the experts in the past the excellent work done by difference is large or due to excessive accord. As a result, after the
the implementation of the GO model is logarithmic power to test timing, minimum preparation failed (60% in ready state index information) for
in the old Goel-Okumoto SRGM by as needed. The time change is applied the selected load display, approval (approval information index 20%
to NHPP model, free coefficient (ie time "t") is logarithmic. it will work coverage) to ensure complete model for the new information. No). Box 1
with the Executive power test "(t)" instead. shows the mutual recognition system as a model to select the correct
load.
3.2. Checking validity of the model For information on two indexes, approved failure rate in Trial-1 are
high. Therefore, the future of information the be expressed more
The rational programming accessible disappointment information appropriately. For Trial-2 load, prepare and approve the error is very
index to evaluate the proposed model show. Before processing ANN, the small, so the model is also suitable for future information.
information collection should be normalized in the range [0,1]. Table 1
shows the information about the product to gather information disap­ 4. Result and discussion
pointment. Adaptation of the model (a GoF) using the Mean Square
Error (MSE) proposed to measure and verify. MSE for the contrast be­ The Once described above solves the appropriate load on the
tween the square of the actual value and the evaluation value quanti­ recommendation model, then the time will use the remaining 20% of the
zation. Moderate MSE compatible display fewer failures and better test information to try to run the model to confirm the selected load.
execution. In order to test the overall performance of the test work will be
prepared for a power of TEF with Wyble testing comparing the current
proposed [11]. Both of [1] In the G-O model. The results confirmed the
3.3. Estimation using artificial neural network (ANN)
execution logic power testing rationality in Table 2, all of which are
reasonably determine the TEF.
The Using feed forward ANN and future computing model to eval­
uate the proposed boundary. Thus, the proposed logarithmic power
4.1. Select appropriate weights of the proposed model
capacity SRGM TEF average value is as follows:
( )
Y(t) = w4 1 − e(− w3(w2 ∗ln(1+t)) (3) Goodness of fit measure of the integrity of historical information
indicates that the nature of fitting. The goal is not only to improve the
Wherein w1, 2, w3 w4 programming reliability model and load, and adaptability of past information, but also to ensure that the model can
the ANN analytical quality. Here, the initial elements of the ANN is describe the future of information. In general, assuming the current and
created by average volume and testing of selected SRGM [16]. The long-time reliability model of legitimacy vision to ensure that the model
programming information may be disappointed (tn, wn, yn) visit to is well depicts the future of information. Application mutually approved
assess weight value. Which is a fitting time tn estimated total test time, method retained (which is usually one of the machine learning process),
for example, months and hours. In hours, yn is related to the total in order to improve the adaptability of past information and the reasons
number of disappointed. described in forward-looking information in the future.
The using ANN, a large number of load placement can also promote a
3.4. Machine learning (ML) technique reliable match. The indiscriminately assigned to the initial mass of the
load, excellent characteristics can be realized. Since the model may not
Consistency of quality consistency of measurement indicating the accurately depict the future of information in a similar manner, and
nature of past information. The goal is not only to improve the adapt­ therefore rely on the least well prepared to load the wrong decision may
ability of past information, but also to ensure that the model can be wrong. Choose to load the preparatory work less, but the number of
describe the future of information. In general, assuming the current and approved fail many, there may be major changes or excessive
long-time reliability model of legitimacy vision to ensure that the model compliance.
is well depicts the future of information. Another approved method (one Then, after the show ready for the selected minimum load failure
of the traditional machine learning strategies) to ensure consistency of (60% ready information index), approval (approval collected 20% of the
information and a description of the future vision of the legitimacy of coverage information), to ensure that the model is well adapted to the
past information and applications reserved. Increase new information. No). Box 2 shows another method of selecting show
When using the ANN, a different arrangement may also facilitate the Fig. 3 an appropriate recognition model for the load.
reliable load matching. According to initial mass inadvertently given The preparation of the two initial weights of the proposed model of
load, it may achieve a variety of excellent cooperation. Load choice heavy sets and mutual recognition of the mean square error assessment.
depends on minimizing preparation errors, Although for Trial-1 and Trial-2, the preparation is almost the same
Spoofing is because the model may not accurately depict the future error, but it is easy to see that the approval of Trial-1 high overall failure
of information in a similar manner. The choose to reduce the load of the rate of these two pieces of information collected. Therefore, the future of
failure of the preparatory work, but if a lot of approval error, the information the expressed more appropriately. In Trial-2 load, prepare
and approve the error is very small, so the model is also suitable for
Table 1 future information. Appeared on
Software failure data sets.
Data Description
sets

DS-1 Delivery 1, is a subset of four programming emissions Wood for Tandem

Computers Company references. After 20 weeks (tn) trial found that 100
programming disappointing (yn), and consumes 10,000 CPU hours (wn). Table 2
DS-2 Delivery from the series computer 2. Try the 19 weeks (tn), goal found
The MSE determined with test information.
120 disappointing Programming (yn), and ran out of 10,272 CPU hours
(wn). Models MSE
DS-3 Delivery 3 from the series computer. It tried 12 weeks (tn), found 61
DS-1 DS-2 DS-3 DS-4
programming disappointed (yn), and depleted 5053 CPU hours (wn).
DS-4 After 35 weeks of Brooks and Motley test () 1301 and disappointing (yn), G-O SRGM with Weibull TEF [11] 0.5327 0.5262 0.4028 0.8935
method found 1846.92 CPU hour (wn) runs out Proposed G-O SRGM with log-power TEF 0.4823 0.4741 0.3485 0.4031

3
L. Li Microprocessors and Microsystems 80 (2021) 103538

Box 1
Selecting the appropriate weights weight value learning machine described
mutual authentication procedure.
1 Prepare a 60% of the index information training. 2 To change the load caused by
errors throughout the organization and use of feed before ANN growth accounting
background, ready to calculate the accuracy of the information index. 3 To verify
the following information approved by 20% of the index. 4 If the approved limit
accuracy is met, the preparation will be terminated. Otherwise, continue to prepare,
until it reaches the limit so far approved accuracy. 5 Were tested by the following
test information index of 20% to confirm the model is determined at an appropriate
load.

Fig. 4. time determination performance.

accuracy of TEF SRGM logarithmic power, and may also be used to

ensure that the discharge time programming. Use ANN boundary
assessment, rather than the usual boundary assessment strategies

Declaration of Competing Interest

We declare that we have no financial and personal relationships with

other people or organizations that can inappropriately influence our
work, there is no professional or other personal interest of any nature or
kind in any product, service and/or company that could be construed as
influencing the position presented in, or the review of, the manuscript.
Fig. 3. MSE for training dataset error rate.

Acknowledgement
4.2. Use of proposed SRGM
This paper was supported by (1) Hubei Provincial Department of
The product testing and delivery for operational use is one of the Education, Research on NHPP Software Reliability Growth Model of
purposes of software reliability growth model. Convey the assurance of Fault Correction, B2016225; (2) Ministry of Education, Science and
discharge time, they use the most objective measure of disappointment Technology Development Center, the Construction and Research of
as, instead of unwavering quality, because there is no integration relies Teaching Support Service System from the Perspective of Wisdom Ed­
accident reliability, the ideal delivery assessment they will receive a ucation, 2019ITA03005.
similar method to determine when to use the model proposed end of the
test. 3 shows a system block down the discharge time value is completed References
programming.
The proposed based on the ability of the TEF number of SRGM [1] A.L. Goel, K. Okumoto, Time-dependent error-detection rate model for software
disappointing conditions are as follows: reliability and other perforinanceineasures, IEEE Transactions on Reliability 25 (3)
(1979) 206–211.
ℵ(t) = dy/dt(w4 [(1 − e](( − w3(w2 ∗ ln(1 + t)))))/dt (4) [2] S. Yamada, S. Osaki, Software Reliability Growth Modeling: Models and
Applications, IEEE Transactions on Software Engineering SE- I I (1 2) (1985)
Set weekly disappointed objective indicators to the preparation of 1431–1437.
[3] P.K. Kapur, R.B. Garg, A software reliability growth model for an error removal
disappointment information index DS-4. Disappointed objective force phenomenon, Software Engineering Journal 7 (4) (1992) 291–294.
has been achieved, the test can be stopped at 25 weeks shown in Fig. 3. [4] A.L. Goel, Software reliability models: assumptions, limitations, and applicability,
Show the as Fig. 4. Based depletion SRGM use, determine the ideal I FFF Transactions on Software Engineering I I (1 2) (1985) 1411–1423.
[5] S. Ramasamy, G. Govindasainy, Generalized exponential Poisson model for
test time (TOPT), but also to determine the firm needed to achieve the
software reliability growth, International Journal of Performability Engineering 2
desired goal of quality depletion. (3) (2006) 291–301.
[6] R. Subburaj, G. Gopal, P.K. Kapur, A software reliability growth model for Vital
Qualit y Metrics, South African Journal of Industrial Engineering I $ (2) (2007)
5. Conclusion
93–108.
[7] S. Ramasainy, C. Govindasainy, A software reliability growth model addressing
Time-sensitive software reliability growth model (SRGM) acknowl­ learning, Journal of Applied Statistics 35 (9- 10) (2008) 1151–1168.
edged that over time, test work has been steady, sometimes it is absurd. [8] R. Subburaj, G. Gopal, P.K. Kapur, A software reliability growth model for
estimating debugging and the learning indices, International Jounal of
Based SRGM exercise smarter, more fit body. As a result, several SRGM Performability Engineering S (5) (2012) 539–549.
previously been proposed to test athletic performance. Because testing is [9] S. Ramasamy, A.M.J. Muthu Kumaran, Dynamically weighted combination of fault
legally unlimited and infinite testing time, so method propose a number - Based Software Reliability Growth Models, Indian Journal of Science and
Technology 9 (22) (2016). Article lD 93967.
of power TEF, it is unlimited testing tasks. The proposed based on the [10] S. Ramasamy, C.A.S. Deiva Preetha, Dynamically weighted combination model for
powers of the TEF SRGM also depicts various information disappointing. describing inconsistent failure data of software projects, Indian Jour nal of Science
Indicate the nature of the consistency of quality consistency of past in­ and Technology 9 (35) (2016).
[11] S. Yamada, J. Hishitani, S. Osaki, Software-reliability growth with a Weibull test-
formation. Method do not guarantee future information will also apply. effort: a model and application, IEEE Transactions on Reliability 42 (1) (1993)
Therefore, use AI strategies to determine the appropriate load, and 100–105.
chosen to represent the past and future disappointment SRGM. The [12] C.-Y. Huang, S.-Y. Kuo, M.R. Lyu, An assessment of testing-effon dependent
software reliability growth models, IEEE Transactions on Reliabilit y 56 (2) (2007)
study found that a better than conventional TEF boundary estimation
pp. I 98—2 I I.

4
L. Li Microprocessors and Microsystems 80 (2021) 103538

[13] P.K. Kapur, D.N. Coswami, A. Hardhan, 0. Singh, Flexible software reliability [20] Soinasundaram Kasiviswanathan, Dillibabu Ramalingam, Development and
growth model with testing effort dependent learning process, Applied application of user review qualit y model for embedded system, Microprocessors
Mathematical Modelling 32 (7) (2008) 1298–1307. and Microsystems 74 (2020). I030293SSN 0141 -93.
[14] J. Xu, S. Yao, Software reliability growth model with partial differential equation
for various debugging processes, Mathematical Problems in Engineering (2016),
2476554. Art. lD 2476584.
Liya Li is a PHD candidate working on Central China Normal
[15] j P.Ro y, G.S. Mahapatra, P. Rani, S.K. Pandey, K.N. Dey, Robust feed forward and
recurrent neural network based dynamic weighted combination models for University and Hubei University of Education. His research
software reliabilit y prediction, Applied Soft Computing 22 (2014) 629–637. interests include biostatistics and statistics, more than 2 papers
[16] Y.-S. Su, C.-Y. Huang, Neural-network-based approaches for software reliabilit y published.
estimation using dynamic weighted combinational models, Journal of Systems and
Software 80 (4) (2007) 606–615.
[17] I. Lakshinanan, S. Ramasainy, An Artificial N euralN elwork Approach to Software
Reliability Growth Modeling, in: Proceedings of the 3rd 1ntemational Conference
on Recent Trends in Computing, lC RTC 2015, March 2015, pp. 695–702.
[18] S. Ramasainy, I. Lakshmanan, Application of artificial neural network for software
reliability growth modeling with testing effort, Indian Journal of Science and
Technology 9 (29) (2016) 90093.
[19] A.M.fl.Muthu Kurnaran, Subburaj Ramasamy, A dynarnically weighted discrete
combination model for all releases of a software system, Microprocessors and
Microsystems 79 (2020). I032903SSN 0141 -933 I.