# Full text of "2 67 1586260060 92 IJMPERDAPR 202092"

## See other formats

International Journal of Mechanical and Production Engineering Research and Deyelopment (IJMPERD) ISSN(P): 2249-6890; ISSN(E): 2249-8001 Vol. 10, Issue 2, Apr 2020,931-954 © TJPRC Pvt. Ltd. ARTIFICIAL POKA-YOKE INCREASES PROCESS RELIABILITY VIA A HYBRID OF THE NEURAL NETWORK WITH ARIMA RESULTS AHMED M. ABED 1 , SAMIA ELATTAR 2 , TAMER S. GAAFAR 3 & FADWA ALROWAIS 4 1 Department ofIndustrial Engineering, Zagazig University, Egypt I,2 Department ofIndustrial Engineering, AIET, Alex, Egypt 2 Department ofIndustrial and Systems Engineering in College of Engineering, Princess Nourah bint Abdulrahman University, Riyadh, KSA 3 Department of Computer and systems Engineering, Zagazig University, Egypt 4 Department of Computer Sciences in College of Computer and Information Sciences, Princess Nourah bint Abdulrahman University, Riyadh, KSA ABSTRACT A product 's sustainability fails if its standard function deviates from the desired working conditions (i.e., by making an unscheduled stop), which makes less competitive. This study aims to develop a smart autonomation system via closer monitoring of the time series' forecasting model ofdeviation behavior to enable it to stopfor maintenance when needed. The proposed methodology (Artificial Poka-yoke; APY) achieves the objective via two sequential phases focusing on improving the operation 's performance, (deviation behavior to stop the machine before antagonhing HBUs essential elements). The methodology alters the ARIMA's trajectory and its error, as a Neural-Network (NN) inputs model is used to simulate them in the second phase to enhance its accuracy by comparing minimhed MAE and RMSE values, through 72 trials and revealing the Poka-yoke index error. An APY approach is adopted for an electrical combustion engine to enhance its reliability level, based on intake control, unloaders valve and rifle drilled. The proposed methodology demonstrates its ability toforecast energy consumption related to that actually generated effectively and accurately. The methodology guarantees a reduction in downtime and waste by 0.71 % as a profit saving via ARIMA (2,1,1) x(0,l,l)6 supported with a neural network to create smart operation. KEYWORDS: Lean Manufacturing, Poka-Yoke, ARIMA, Neural Network Model & Deviation Forecasting Received: Feb 22, 2020; Accepted: Mar 12, 2020; Published: Apr 06, 2020; Paper Id.: IJMPERDAPR202092 1. INTRODUCTION Deep-Lean is a modern interdisciplinary concept which aims to eliminate the waste at the sub-causes level during manufacturing or services related to the TBL (Triple Bottom Line) elements and adopts the motto “environment, machine and people' are friends”. It also aims to evaluate and manage the performance of operations [2] and to increase reliability, availability and reducing environmental risk via reducing fuel consumption. In Egypt, the governorate objects to increase its exports ad satisfy customer via growing interest in the reliability concept, according to which the process is given the ability to stop working if its output will harm one of the TBL elements, thus focusing on increasing the processes’ ability to detect mistakes and correct them immediately. Poka-yoke (i.e., reliability philosophy) is a mistake-proofing approach proposed by Shigeo Shingo to enhance the reliability concept, which aims to eliminate any problem created by deviating from product working conditions, by preventing or correcting mistakes as early as possible. Every working span of products' process is challenged with TRAN5 STELLAR • Journal Publications • Research Consultancy www.tjprc.ors SCOPUS Indexed Journal editor@tjprc. org Original Article 932 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais many errors incidence at any time [17], using Poka-yoke, the deviation is moved toward the rejection zone under working conditions, avoiding harm to any of TBL elements. Since the end of the 20th century, the reduction of fliel consumption has been at the top of every manufacturers’ agenda, via control the hmction faulty carburetor valves m-of any combustion engine [26] (i.e., used in vehicles, steams, airplane ... and so on). The fuel consumption will be reduced by controlling carburetor’s valve actuation behavior, that prohibits the fuel stream in its tracks, as a perface to drainage through engine shutting [34]. This will help in maintaining the reliability of this valve during its working life with minimum deviation from standard specifications. Therefore, an interdisciplinary approach advocates contributing to the sustainability of TBL keep down, with good hmction via its design improvement of these valves to be interactive and subjected to kaizen [32] to reduce pollution. Poka-yoke is a technique that aims to reliably monitor the deviations from standard limits of working conditions over a long time period, which calls for developing an accurate methodology for prediction of malhmction. In the case of a combustion engine, it is extremely difficult to achieve a zero-leakage consumption rate while working in a large environment that many manufacturing hmctions for every device's components. The units inhuence the hiel consumption the of the combustion engine shown in Figure-1: the motor, compressor/cylinder unit and valve unit that are all interested in check the methodology of the proposed prediction process. Ongoing monitoring becomes viable by predicting process deviation [6] and solutions can be discovered for process reliability through analysis of the malhmctions. Figure 1: Hydraulic Generator Components. The monitoring procedures are carried out by measuring specihc hmctional factors modeled with ARIMA supported with NN to enhance reliability level. The operation's C P is a popular level parameter that explains the reliability standard, but this is not enough in cases that include human interventions [1]. Thus, the use of the Poka-yoke index is suggested to evaluate the reliability in effect on the process on the social, economic and environmental aspects together (i.e. the TBL elements). Any deviation during working operation must be addressed by stopping the process until maintenance done. Shingo suggests three Poka-yoke categories: judgmental, informative and at source [4]. Poka-yoke sorts and prods the defective products that are deviating from the acceptable standards, without going into working methods to tackle the causes of their defects. Informative Poka-yoke collects feedback information on the defects (i.e., malfunction) to reduce their rates gradually by improving the manufacturing method [34]. Informative Poka-yoke includes sampling checks in SQC/SPC. The three types seek information about lifetime reliability, which is the working time of every part in the process without deviation from standard working condition, CL. Therefore, it is necessary for researchers to concentrate on waste sub-causes, particularly with regard to deviation from standards, which is called the Deep-lean level of artificial Poka-yoke framework, as illustrated below in Figure 6.1. The device components explained Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results 933 in Figure 1 made up Piston moving forth and back in a cylinder, with discharge and suction valves and lead to increased fuel consumption in function failure [6] due to bad in a hole’s radius or thickness dimensions. The discharge and suction valves close and open according to pressure variations among the cylinder and entry or exit manifolds respectively. Poka- yoke is significant case in superior schedule of investments. Artificial Poka-Yoke (APY) is a suggested methodology that guarantees a long reliability lifetime, by stopping working and correcting the deviation if it is possible or changing the part to ensure the safety of TBL elements [14, 26]. The deviation has non-linear behavior. Therefore, in this research, the reliability lifetime was forecast via ARIMA aided with NN (which monitors a relative risk of an output by studying the significant parameters of its sub-causes at the minor level of lean implementation) for measuring the amount of deviation created during working. ARIMA, created by Box and Jenkins in 1970 [9] was based on observations created from first day of working and consists of a dataset which can be used to forecast the deviation behavior for creating effective control in a specific working span and time based on an APY framework, which has the advantage of improving the prediction accuracy (Luxhoj, et al., 1996 [22]; Balkin&Ord 2000; Medeiros &Veiga, 2000; Tseng et al., 2002 [35]; Zhang, 2014 [46]; Chen & Wang 2007 [12]; Isinkaye et al., 2015 [19];Qiu& Song, 2016 [31]). 1.1 ARIMA Modeling Approach Needed in Phase-1 The ARIMA (p, d, q) forecasting model of fuel consumption can be represented by general terms, as follows: yt = p + -i- ™ -i- a P y P -2 + 1 -V<j-l (i) where p is the autoregressive number and q is the number of lagged forecast errors in the prediction equation, based on plots of the AutoCorrelation Function (ACF) and Partial AutoCorrelation Function (PACF) [29]. 1.2 Artificial Neural Network Modeling needed in phase-2 The (NN) model is characterized by emulating the nonlinear or linear data behavior accurately [7]. The NN is determined by the characteristics of this data fed forward from ARIMA and DOE optimizer outputs to work through some of the hidden layer related to time series, which is formulated as follows: q P Yt = ^ w j * 3* ( w oj + ^ w a * x t.d + s t i = i (1.1) Where wij(i= o,i, 2 ,..., P ) and wj(j= 0 , 1 , 2 ,...^) are the connection weight parameters. The terms p, q, st, wO and w0,j and g x are the number of input nodes, the number of hidden nodes, error term and g x = x _*_ r , while the time series model for linear and non-linear output data, formulated as y t = l “ H- nV t . where V t is linear, while nl t is nonlinear data output from ARIMA. APY consists of two sequential phases; phase-1 is to obtain a linear model from ARIMA, while the error obtained from phase-1 behaves as nonlinear, e r = > p t +3?t. Therefore, the second phase (phase-2) is created to tackle the nonlinear predictable and its error via the NN model follows t-i ■■■ where/is a nonlinear function, and cot is a random weight variable. The APY forecasting model would then be written as: www.tiprc.ors SCOPUS Indexed Journal editor@tjprc. org 934 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais yF t =Cl + ?OF f . ( 2 ) The accuracy of the reliability is based on both linear and nonlinear characteristics’ behavior for the available series of deviations. Two key performance indicators are used to test the accuracy: the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), which are calculated to compare the results’ accuracy, as follows: MAE = 2l=i y. -y f r n (3) RMSE = j ra- )S(y t -yF t y ^ t=1 /n (4) Each dataset collected concentrates on a specific working condition to reflect its deviation behavior [6, 18]. The proposed APY model outperforms current models, because it forecasts by giving the amount consumption (in terms of its deviation) [10, 17], and whether this was acceptable or not. 2. COMBUSTION ENGINE FUNCTIONALISM The experimental study concentrated on; 1} the valve plates as a cause fuel consumption instability, and 2) The rotary valve [2]. The analysis presented here matches to an indoor combustion engine of a cylinder space 6.64 cm3, active with R134a and a nominal frequency of 50 Hz. Suction, pressure and discharge operations are the central operations of the selected device, which are carried out during valve plate slots and discharge slots [2] 2.1. Experimental Design This study is aimed to increase the reliability of the processes to avoid extra fuel consumption and back off the torque power, if worked in poor condition. The experiments were focused on valve unit of intake and Safe-Q-lube unloaders and cast-iron valve seats, which included the valve plates, valve gaskets, cylinder head and muffler. Figure 2: Schematic Valve Unit of Generator. Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results 935 Figure 3: Discharge Orifice of Valve Plate and Crankcase. The selected control factors (i.e. tolerance limit) for valve units are the valve thickness (V th ; 2.45 #fll f m :3...S5 discharge orifice radius of the valve plate (R l?p ; 3.S1 3.920 ^ n mm) and discharge -•? 7 u in -•=? 7 u in orifice radius of the crank case as significant factors affecting the process's reliability (IR^^; 2.275525 as illustrated in Figure 2 and Figure 3. Any deviations in these limits that cause extra fuel consumption, which is measured inmm 3 xl0 3 /sec, and reduce its torque are recorded in Table 2. The experimental study handled the real data of 24 combustion engines, based on eight samples for a two-level set of DOE 2 k /3 replicates. All samples were produced and examined maintaining the actual manufacturing environment the remnant of the parameters of other parts. Figure 4 illustrates the important factors that have to be planned though valve plate or crankcase processing to conserve the fuel consumption (<0.2 mm 3 xl0 3 /sec). Figure 5 suggests that the valve thickness should be controlled at 2J5mm , and the discharge orifice of the valve plate radius should be 3.0 mm , whiles the discharge orifice of the crankcase should be 2.525 mm (all these values with tolerance in mm). The operational effectiveness is based on four key factors [rotational speed, altitude path, fuel feed rate and thickness of valve]; the rotational speed which is measured in rpm, is recommended to be constant, and its effect on the average altitude path of 21.4 mpm , which is matched with [457:459 rpm], whereas the fuel feed rate [8:12 cm 3 /sec\ is based on the altitude path length (second factor), and the thickness of the valve. The Pareto chart cleared in Figure 4-1 explains that variations build on the way that rotational speed intersects with feed and thickness. Monitoring the operational deviation vs. time, for every significant factor, as shown in Figure 5; fuel to be conserved and be prevented for the extra fuel consumption. Pareto Chart of the Effects (response ts Gas evaporative emissions, Alpha = 0.03) 0.2054 F«!*r Hirint A v^lve thickness [irnn] e Dlscharga oriflce R of crankcas c Djscharge orificc R of valvs Effect Lenth s PSE = 0.039625 Figure 4: The Important Factors cause Yiolate Euler's Equation. www.tiprc.ors SCOPUS Indexed Journal editor@tjprc. org 936 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais Pareto Chart of the Effects (respor»se ts FLinctionial de^iatlon, Alpha = 0.05) 5.49 Lenth's PSE — 1.A5B75 F*ctor Ntmt A Revolvable Speed[rpm] R Altitude path[mrri] c Puel Feed[m3] D Thickness of ^elweCmnn] Figure 4.1: The Important Factors cause the Functionalism Deyiation. 3. PROPOSED [APY] FRAMEWORK The proposed framework detailed in Figure 6.1 focuses on predicting the deviation during the working period using ARIMA, then feeding the dataset with expected values of its errors in nonlinear behavior based on the NN model, to increase the accuracy of predicting its reliability. The leakage rate, based on a Poka-yoke index derived from Euler's turbine equation based on, where the power output collected in Table 2, P — = copQ (r lv _q ifl Cosp iri — r {mt q {mt Casp cut ') as illustrated in Figure 6 and PYi =/(tp,^), which is a function of the consumption amount *P G p required (¥ = 53 e + E ), where - is the deviation, appears in Figures 5 and 5-1 (£ = 1 — d) and the output feed NN as a function of *P (NN=/(^)), whereas the second parameter (i?)is merely a dual parameter (among zero and unit) that measures the importance of working induenced according to the integrated DOE equations. Higher rates of i? increase the variations and command to a major stage of leakage occurring, and lower rates lead to the opposite, and are formulated as Eq. (5) and Eq. (5.1) to fade any trend in malhmction behavior for each of the significance factors shown in Figures 4, 4-1 and 6-1 and determine the best working. The parameters shared in the leakage, rates that appeared, as illustrated in Figure 6.2 according to PYi=f(significant limits (k), uptime, lifetime) and its significant factors =/[(Temperature, fluid flow rate, fluid velocity, fluid density)ltarget] and were calculated according Eq.5 for fuel [6.8:8 mm 3 / sec], for working factors to be [85:115 mm\ and for valves to be [0:30xl0 3 mm]. PY^t + St) =f(<p) = ln a^js) = \ 1 (1 -ti)if> s (t) + —) A t/y(£) (1,'^A^r (t) - zif> s (t)) Z Ij—I S ’■-£ (1 - w8t) + dE s (t) (5) (5.1) where A is the discrete Laplacian for difference between consumption at t and t-1 period and pt is the mean value for z significant factors. The variations for 6 lag [34] to dissolve the way, as explained in Figures 14, 15 and 16, the datum Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of 937 the Neural Network with Arima Results composed have to be normalized through rolled up and squared I logarithmic to trial its impact with time variations to stop at a certain time for protecting TBL elements. Figure 5: The Recommended Rates of Important Factors that Reduce the Fuel Consumption. Figure 5.1: The Recommended Rates of Important Factors that Besiege the Deyiation. Table 1: The Fuel Consumption with mm 3 /sec for 33 Months Mcmths period^ Ml2 Ml6 Ml9 M 23 M 25 M 27 M 28 M 29 M 30 M 31 M 32 M 33 Os Us Us Os ds <7s Us Us Us <7s <Is <Ts 1 79.47 59.31 94.24 117.02 133.54 129.78 146.72 139.4 153.95 128.47 135.54 119.92 2 69 59.34 97.43 116.89 123.42 144.22 124.51 134.23 134.57 142.39 145.7 136.52 3 60.97 70.32 117.65 84.39 126.33 141.11 126.22 138.74 122.67 140.95 139.19 155.46 4 49.46 70.07 120.14 82.3 127.8 142.25 124.04 128.79 122.79 138.69 125.6 142 5 70 72.38 120.29 82.46 128.02 124.99 145.2 143.85 129.32 148.48 127.35 114.81 6 66.41 74.39 123.1 84.83 125.95 136.72 124.35 124.45 127.09 128.17 129.85 136.27 The local dataset adopts a neural-network model and considers the second phase of the proposed conceptual framework APY, as illustrated in Figure 6-1 and defined in Table 2. Table 1 illustrates one of the outputs measured and expresses the fluid_flow_rate mm 3 l sec and other fuel consumption/sec. Figure-7, which records some of the statistical frequencies [15] of deviation values, a s (deviation), if the important factors explained in Figure 4-1 are neglected. The a s raised the error proceeding (i.e., the variation is reAected as a mirror for malfunction in the form of backward and forth, movements and burning) and k expresses the span of variation (significant limits according to the Pareto chart), which have www.tiprc.ors SCOPUS Indexed Journal editor@tjprc. org 938 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais maximum probability of deviation appearing. Euler’s Turbine Equation Q = Fluid flow rate p = Fluid density q = Fluid velocrty Define Cujtom Factorial Dejign • Low/High Low and High Values for Factors / 1 A p = Incidence angle / v y/ioAN^y ^ V = Tangential tluid velocity i \ A V = q cos P V %/ j r = Turbineradius cj = Turbine rotational speed T = Torque P = Power output Factor Name Type Low High A fkiddffB*y Numerrc ®l 349 B fbdflowrat Numenc 1.9% 2.480 C Torque Numenc *| 2.5 12.0 D Temperature Numenc jJ 35 49 E fkjid vetoaty Numenc ▼] 2.55 36.50 Torque T= pQ(r«Vin - romVout) Power P = coT = copQ(rinqin cosPin - romqout cosPout) Figure 6: Euler Figure 6: Euler's Turbine Equation’s The data introduce the variation's span range of 72 trials over 33 uptime months (i.e., the part must be replaced by newest one) with (a s ; rara), as shown in Table 2 (Location A). As stated in Figures-3 and 7, the medium variation in fuel consumption is 113,414rara 3 /sec, where the minimum was 69x10 3 rara 3 /sec in a month (12) and the maximum was 155,46xl0 3 rara 3 /sec in the month (33), which increased by 0.73% and deviated by about m by 24,08454xl0 3 rara 3 /sec (leading to leakage in the part's hmction), the accepted variation range for the fuel consumption is 0 .16<PY X < 1.56xl0 3 . These rates indicate the Heterogeneity of collected data, but according to the T-test, it clarified normal distribution. The data were analyzed at other locations (i.e., B, C and D), as illustrated in Figure 7. oi beginning of woik Error prerention, iinmediate error detection and coirection before hann TBL elements In downstream process sia Neural- Figure 6.1: The APY Conceptual Framework. Table 2: APY (APY) Cycling Functions Appeared in Figure 6.1 NO. Action 1 Inspection of fuel consumption sensors especially forfluid flow rateandfluid density 2 Check matched within standard limits control 3 Determine the leakage rate for fuels per sec 4 Check for variation span a s Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results 939 5 Create the central dataset for o s for every fuel output value and expect via ARIMA the future value with time series 6 DB for validate the process to continuo or invalid and make query 7 Send Query to APY methodology in central database based on ARIMA 8 Check stationary if the a s < 0.01 (functional analysis) or not 9 Update the input information with error value 10 Analysis the ARIMA and its error via Neural Network 11 If the observation not stationary, Delete the fade of the trend observed 12 Forecasting the next consumption via Neural-Network 13 Send forecasted a s in mm for APY methodology in central dataset (x 3 ) 14 APY approval if a s <0.01 hom central dataset to stop for rework 15 APY action; [Send data of stopping (if a s >0.01) 1 continuo otherwise] for the three signihcant outputs and calculate the process capability 16 Count the generator downtime and reliability value 17 Continuo the APY cycle The same data table was collected for all the DOE's significant factors illustrated in Figures 4, 4-1, 6-1, through three different sensors to feed the dataset with data used in forecasting deyiation behayior. Pareto Chart of the Staodardized EFFects (response is Power[KwatVhl A^pha - 0,05) 1.997 Standardized Effect Figure 6.2: Pareto Chart of Gained Power. Pareto Chart oFthe Standardized Effects (response rs Fuel Coreu mptior, Alpha = 0-0S) 1.997 E I anrlard ize dl E ftect Figure 6.3: Pareto Chart of Fuel Consumption. www.tiprc.ors SCOPUS Indexed Journal editor@tjprc. org 940 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais q O O D -T* O 7T n J o m a> &> 0TQ m ^n> < < (D 5 5 5 8 O J m cd 0Q Figure 7: Sensors Identification. Contour Plots of Power[Kwatt/h] a o ■ >00 200 300 )00 200 300 >00 200 300 I tiii >00 200 100 2 0 U 1A 20 22 1.4 20 22 2-4 H a a=n ■ >00 200 KO B 20 12 U s n i s n Pow«r[Kw8ttyh] < 5 ■ 5 10 ■ 10 15 ■ 15 20 ■ 20 2S ■ 25 Hold Valu« flijd denaty 87 flud flow rite 1.8% Torque 2.5 Temperjhjre 35 turbine raduis 4 flud vekxity 255 Figure 7.1: Analysis of A, B, C, D Surfaces appeared in Figure 7 via NN 4. PROPOSED APY IMPLEMENTATION The proposed algorithm [APY] observes the deviations of the PY index at significant locations (i.e., A, B, C or D) during working time via specific sensors. The following leaking areas were simulated through Minitab. The first track in the proposed framework was intended to analyse all significant factors that directly affect the product hmctions and provide immediate error detection to prevent toxic fuel consumption. The data shown in Table 2 is drawn as explained in Figure-8 to identify the characteristics of the deviation's behavior and to control it via NN, where it reveals many periodic Auctuations that have trends approximately out of standard specification vs. time, which revealed seasonal properties (natural effects of product usage). Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results 941 The deviatiori rneasures for different locations on OSS Figure 8: Main Characteristics of Yariation Distribution. Figure 9: Autocorrelation of Yariation Distribution. The data shown in Figures 8 to 10 represent the cumulative summation of back-and-forth ward responses that control the deviation's trend [19]. It is essential to extract the ACF and PACF for the variation distribution, as shown in Figures 9 andlO respectively, which reveal the parameters are significantly different around zero, to pin at #18, and are out of the confidence region, which must be within —0.215 < ^ <0.215. Therefore, a stationary case test is mandatory. 4.1 Stationary test This test tackles the data using the logarithmic (logio) to fade the trend, which appeared to accurately forecast the output, as shown in Figure-8 and Figure-ll respectively, for the variation behavior that is essential to feed the neural network model later, and presents a close view of the research objective. The ARIMA chooses parameters by extracting ACF and PACF, which are significantly different around zero until they spike (13) and not within confidence range —0.215 < ^ <0.215. L Iung & Box states the refusal of the main hypothesis Ho in order to Q»stat = LBQ (TBF) = 383.591 > = 34.83, Now refuse B®:p L = pn = AA= p% = 0 and faith of no fixed in the variation distribution data. www.tiprc.ors SCOPUS Indexed Journal editor@tjprc. org 942 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais 4.2 Disappear the Trend Select the near distribution which decreases the trend and echo the differences for its value until this trendfades -— ^t-LdoguO as illustrated in Figure-12, but the LBQ must be repeated to the first difference to be LBQ ( TBFf irst = 32.52 < * 2 ri 7 F o 1 oO = 33.41: Therefore, we have not taken the second difference for the TBF data and accept Ho. T locj of lon »» irc**» for di fferent locotlons <>« » OSS Figure 11: Absence of Trend. Autocorrelatlon Fi»tc t lon for Flrst_Dl f f ( loc|(Devlat lon Vdlue) (wrth S% sQnifk;ance 1 1 mlts fbr th»e autDcorrelataons) Figure 12: ACF after First Difference. Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results P<«rticjl Auto Correlation of deviation distribution in different locations .-2 5 !-c tWrJ 1 J I - J- - 2 4 6 8 lO 12 14 16 18 Figure 13: PACF after First Difference. The log of ck*viat ion nK*cisurc*s for different locations on OSS Figure 14: Absence of Trend. Figure 15: ACF Seasonality First Difference. PACF of <k*viatlon dlstrlbution In locatlons after fade perlodlcally effect Figure 16: PACF Seasonality First Difference. 943 www.twrc.ors SCOPUS Indexed Journal editor@tjprc. org 944 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais The TBF should be in equivalent to the center of the joints, to prove the absence of the trend, as mentioned in Figure 13. Then, preferably, [d=l], but Figure 12 and Figure 13 explain the seasonal behavior (which appeared for repeated maintenance actions) for the deviation at the location of spike 12. Therefore, the focus was on fading the seasonal attributes and determining the critical span that has maximum probability of malhmction (unreliability) [17] the through time scale : LBQ (TBFf irst wesimGUty d\ff) = 20.63 < = 30.58. Therefore, accept the assumption H 0 for the seasonal case. If the AFC spreads down toward zero by a power sequence, then the rank of (p) is determined through number of AFC, which is different about zero. However, if the PAFC moves down toward zero by the power sequence, then the rank of (q) is determined through the number of statistical correlation (MA) [33]. Figure 17: Residual Plots for Seasonality Deyiation Behayior. Time Series Plot for Deviation Value(maicometer) (wlth tbrecasts and thelr 95% confldence llmlts) Figure 18: The Prediction Phase for the Deviation Distribution. ** Convergence not met after 26 iterations ** Final Estimates of Parameters Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results 945 Type Coef SECoef T P ARl -0.1362 0.1649 -0.83 0.414 MAl 1.0532 0.0097 108.53 0.000 SMA 12 0.9415 0.1383 6.81 0.000 (SMA represents the number of maintenance) Constant 0.46979 0.0183 25.76 0.000 Differencing: 1 regular, 1 seasonal of order 6, # of observations: Original series 66, after differencing 53, Residuals: SS = 5796.49 (back forecasts excluded), MS = 136.99 DF = 52 Modified Box-Pierce (Ljung-Box) Chi-Square statistic Lag 624 36 48 Chi-Square 12.7 14.6 34.6 * DF 8 20 32 * P-Value 0.124 0.801 0.347 * LBQ (T BFf irst \-\ty diff ) = 12.7 < *“ci2youwJ = 26.22 That shows the non-importance and the model is effective to forecast the variation rate. Table 3:Forecasted Deviation for Table-2 data Forecasts the deyiation distribution from period 72 [99% Limits] Period Forecast Lower Upper 73 175.484 157.650 193.318 74 158.606 137.880 179.331 75 154.242 132.400 176.084 76 158.009 135.592 180.427 77 156.718 133.926 179.509 78 156.297 133.215 179.378 79 159.317 135.986 182.649 80 160.406 136.843 183.968 81 152.119 128.336 175.901 82 148.893 124.896 172.890 83 130.424 106.217 154.632 84 147.070 122.655 171.485 www.tiprc.ors SCOPUS Indexed Journal editor@tjprc. org 946 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais o a.oooo low flLJ»Cl dtr [3-19.0] 07 Q —rl- , 2 ,-ieo , (2.0365] i.Meo 2.50 Te nrTSDer-at: o, [49.0] 3S.O 8.490, [8.490] -tV.Hd ve- „36.50, [2.550] 3.550 C o cnp cts. tts Oes m- st» »11 ty 1-0000 PawnrJKw Targ: Z5.0 y “ Ers.o C* — 1.0000 — _ (»> 0.00000 T ennperat s?o flLJld fl c I:3i3] 3L-W60 Torque «Cilci de tLJrbine !d (4%8j [33:83 cl:33S 3 S-SO 33-0 J-3SO Campos ItD De&ta-atoirity 0.00000 Power[Kw Targ: 25.0 y = 12.3462 d — o.eioooo (b> Figure 19: The Optimum Yalue for Reliability and Maintainability. The model checks the optimum values to stop working for the factors illustrated in Figure 19-a, while Figure 19-b proves that any deviation of significant factors fall the y to 12.7 as in Eq.6. Rapid maintenance is required and studies of the deviation in working conditions illustrated in Figure 20 to analysis the PYi. The intake unloaders’ diameters are fixed between [10: 13 mm] or between [20:24 mm] as Safe-Q-Lube suggests, and this achieves the consumption suppression. Figure 20: The Fuel Consumption according to APY. Contour Plot of Power[Kwatt/h] vs fluid velocity; fluid flow rate Figure 20.1: The Optimum Replacing Time according to Significant Factors Output. Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results 947 5. CONCLUSIONS In this study, the ARIMA time series was used to create short-term forecasting for the behavior of deviations for all the significant factors shown in Figures 4, 4-1 and 6-1, for all selected components which may be involved in causing global warming phenomena, especially in Egypt (where TBL is volatile), and the results were enhanced using an NN model that adopted the Poka-yoke principles known as APY methodology to mitigate the downtime vs. overall time. The forecasted output (ARIMA) and then inputs to NN are compared according to MAE and RMSE as illustrated in Table-4. This work contributes to improving the reliability level of any process' conditions and is helpful in rapid intervention to prevent the malhmction event and enhance the reliability and extend the guarantee by 18 additional times and reduce the malhmctions, as illustrated in Figure 21. This paper has presented a general reliability hamework via statistical modeling to control counted data. Table 4: APY Performance using Correlation Coefficient APY perfonnance using MAE and RMSE Amount of Temperature B C F1 ui d_fl o w_r ate miTi 3 /sec. Fluld_veloclty m/sec. Fluid denslty IS7ni‘ Models ]VXAJE RJVISE IVIAF RATSE ]VLAE RJN.TSE TwIAE R]VISE ARJMA 0.S310 1 1472 O 1401 0.2254 7 5302 13.348 1.1815 3 2915 NK 0,6031 3.1 109 0 0640 0.2525 2 4392 3 0302 2.2206 1.S25© APY 0.2285 0.2501 0.0421 0.0869 2.0413 1.8714 1 3733 O 5164 [AJPY perfomiance using correlatioii coefficients correlation coefficient TModels Temperature °C Fluid_now_rate mm 3 /sec. Fluid__velocity m/sec. Eluid deusity FT/m 2 ARIMA 0.373 0.357 0.352 0.631 NN 0.755 0.746 0.937 0.757 APY 0.533 0.506 0.503 0.570 MAE and RMSE reduced percentage error for all parameters APY ParametM-s MAE reduced error RJVISE reduced error ^ Temperature 22.74 10.52 Fluid Jlowrate mm 3 /set. 44.6E 38.02 Fluid velocity m/sec. 64.05 61.81 F1 ui d_d ens ity N/m 2 30 18 32.53 Test Results; the results of testing for ANN used in this work using unseen data are shown in Figure 21. The convergence condition is considered to be achieved when the range between actual values and predicted output is greater than 0.80, referred to as the limitation of the training data set. There were no special deviation events; the APY was based on 72 trials of data and continued as illustrated in Figure 22 to Figure 24. www.tiprc.ors SCOPUS Indexed Journal editor@tjprc. org 948 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais Tra.ning R=0 99905 Validation R=0 96781 Test: R=0 99226 All: R=0 99137 Target Target Figure 21: Training Result of Proposed Network. Fluid flow rate distr!L>Liotlon comparlson Vkrt.bl* -*- [APY] — » - [ARUyiAj Figure 22: Pearson Correlation = 0.152 P-Value = 0.134 for fluid_flow_rate Output. Flt iicl Veloci ty dlstribuation comparison Figure 23: Pearson correlation = 0.371, P-Value = 0.000 for fluid_velocity output Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results Pkiid density distribuation cornparison V«ri*bl« —• [APY]_2 — ■— [ARIMA]1 Figure 24: Pearson Correlation= -0.064, P-Value = 0.532 for fluid_density output Figure 25: Analysis of Poka-Yoke index for Power Generated. Figure 26: The Optimum Consumption and Generated Power. 949 www.tiprc.ors SCOPUS Indexed Journal editor@tjprc. org 950 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais Figure 26.1: The Contour Plot for Fuel Consumption and Power Generated. Thc ubout Ok." vl* AFAJMA offl(rxS Usi-Kj Tr^ft>ruf rrkJltxi WiSt Ldtrribda = O.S v i.wrr? Afrw TrmjRuTTiMlori I.IHH Tlf )F* t.TZ L L L Puoiiul I.WrJ^l Cipibkt Cp 4JJ c«. **i C«J 4M w 4*1 C*p*t-fc» r Pp PP*. 4.4-1 Pfti 4JU Pp* 4.4* Cpm 4.» Otmn4d P«rfof*VTMK£ Dffl V LSL 4JH WW > USL «0333.»» WW TfM*l «3333.»» Eip.Wiih PPM -.LSl* wn - osl* OPM Tm#I 3*i«r&? I . p . O'.uil Piiit.-iiri'ij.ri.'u PPM - L9L* 1 PPM p U5L* i?l-*3 ,M Wl T««1 1 Mtt3.ro (a) ®h°~" mTI ctl iattondbout (he larget usirn:j autonomatlon APY Methodoiogy Using Boot*Cas( Trsnsiormatbn With Lan*da ■ 0.5 Pwm D-tS-a 31 T*bJ« 3«.S USL « £n«ltMiu 3«*rs IJ SflD*v(Wrfw>3 OrG+ress 5tO*Y(CwJ0 ].?AS« All.r Tr^-.i/t.nn L5L* 3.S3S3S f J31H USL* S#T|*M*-n* 3*33 ■Si&ivl.UNV,p) + O.CridEAOS st&*y(0¥*r^ «Mttttt — V#fiki-, — — OwJ Pm*IWuJ I.V.U I.t-:i C j£‘jL,L< y Cp l.-ss u i CPU 1.33 crt ta> Ov*Ai <C*pffbSl1f f»P 4.3S P<X P«» ]h£l PpJ( 91? Cjli, i»J>: 9.6 9.7 9.8 9.9 Ob&KV*d Pwf(HIMIKt PM^LSL OAI P- tM > USL O.Mi E .p. MvVhw) Ptdwrnarwt LSL 4 LL*.lf PPM > LKL* Ci RPM T(Htl IL3.I? Eiff.. PttLiMmarKt pem ]&Ul.4S PPM > USL* 1« K PCM Tm*J VA?t£* <*>) Figure 27: Stages of Implementing APY Testing Methodology. Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results 951 Contour Plot of Power[Kwatt] 2.4 £ 2.3 <Q i. 5 O 5 = ■o 2.2 3 c 2.1 2.0 100 150 200 250 300 fluid density Figure 28: Contour of Power [Kwatt/h]. ACKNOWLEDGMENTS This research was funded by the Deanship of Scientific Research at Princess Nourah bint Abdulrahman University through the Fast-track Research Funding Program. The work reported in this article has been conducted while some of the researchers are affiliated with Princess Nourah bint Abdulrahman University. ConAicts of Interest: The authors declare no conAict of interest. REFERENCES 1. Ahmed M. Abed. Adjust Jidoka Occupational Fatigue to Reduce Idle times using Data Mining as Lean Tool. EIJEST, The Egyptian International-Journal of Engineering Science and Technology2016; Volume 19 (2);pp. 312- 318http://www. eijest. zu. edu. eg/index.php/EIJEST/issue/view/94 2. Tan M, Tan L, Dara S, Mayuex C. Online defectprediction for imbalanced data. In: Proc. Int’l Conf. on Softw. Eng. (ICSE’13 SEIP), p (To appear), 2015 3. Zhang H, Zhang X. Comments on‘ data mining static code attributes to learn defect predictors’. IEEE Trans SoftwEng2007; Volume 33(9); pp. 635-637 4. Y. B. Wijaya, S. Kom, and T. A. Napitupulu. Stock price prediction: Comparison of Arima and artificial neural network methods—an Indonesia stock’s case. in Proceedings of the 2ndInternational Conference on Advances in Computing, Control and Telecommunication Technologies (ACT ’10), Jakarta, Indonesia, 2010;pp. 176-179 5. B. Johnston, L. B. Maguire, and T. M. MciGinnity. Downstream performance prediction for a manufacturing system using NN and six-sigma improvement techniques. Robotics and Computer-Integrated Manufacturing2009; Volume. 25 (3); pp. 513-521. 6. Antony J., Snee R., and Hoeril R. Lean Six Sigma: Yesterday, Today and Tomorrow. International Journal of Reliability & Reliability Management 2017; Volume 34 (7), pp. 1073- 1093, doi: 10.1108/IJQRM-03-2016-0035. 7. M. M. Mostaja. Forecasting stock exchange movements using neural networks: empirical evidence from Kuwait. Expert Systems with Applications 2010;Volume.37 (9); pp.6302-6309. 8. Bomoisin J., Stark R., Seliger G. Field of Research in Sustainable Manufacturing. In: Stark R, Seliger G., Bonvoisin J. (eds) Sustainable Manufacturing. Sustainable Production, Life Cycle Engineering and Management. Springer, Cham 2017. Power[Kw*tt] - 20 .... 30 Hold Velues Torque 2.5 TempereUre 35 turbne reduis 4 fluid veloc*y 255 www.twrc.ors SCOPUS Indexed Journal editor@tjprc. org 952 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais 9. O. Adewumi and A. Moodley. Comparative results of heuristics for portfolio selection prohlem. in Proceedings of the IEEE Conference on Computational Intelligence for Financial Engineering & Economics (CIFEr ’ 12), pp. 1-6, New York, NY, USA, March 2012. W. C. Chen, G. L., Fu, P.-H. Tai, and W.-J. Deng, Process parameter optimiiation for MIMO plastic injection molding via soft computing. Expert Systems with Applications 2009; Volume 36; pp. 114-1122. 10. Heider, D., Piovoso, M. J. and Gillespie Jr, J. W. Application ofa neural network to improve an automated thermoplastic tow- placement process. Journal ofProcess Control 2002;Volume 12;pp. 101-111. 11. C. Cho, D. D. Kim, J. Kim, D. Lim, and S. Cho. Early prediction ofproduct performance andyield via technology henchmark. IEEE "Custom Integrated Circuits Conference (CICC)"2008; pp. 205-208,. 12. Freitas J. G., Costa H. G., and Ferrazi F. T. Impacts of Lean Six Sigma over Organiiational Sustainability: A Survey Study. Joumal of Cleaner Production2017; Volume 156; pp. 262-275, doi:10.1016/j.jclepro.2017.04.054 13. Gupta V., Narayanamurthy G., and Acharya P. Can Lean Lead to Green? Assessment ofRadial Tyre Manufacturing Processes Using System Dynamics Modeling. Computers and Operations Research2018; Volume 89; pp. 284-306, doi: 10.1016/j. cor.2017.03.015. 14. S. Herhold. Comments on ScottKnottESD in response to "An empirical comparison ofmodel validation techniques for defect prediction models". IEEE Transactions on Sojtware Engineering 2017;Volume 43(11); pp. 1091-1094. 15. K. Aishwarya Raghavi et al. Human Iris Recognition Using Fuzzy Neural Concepts. 2011 International Conference on Bioscience, Biochemistry and Bioinformatics IPCBEE, Volume.5 IACSIT Press, Singapore2011. 16. Y. B. Wijaya, S. Kom, and T. A. Napitupulu. Stock price prediction: Comparison of Arima and artificial neural network methods—an Indonesia stock’s case. in Proceedings of the 2ndInternational Conference on Advances in Computing, Control and Telecommunication Technologies (ACT ’10), pp. 176-179, Jakarta, Indonesia, December 2010. 17. L. Madeyski and M. Jureczko. Which process metricscan significantly improve defect prediction models? anempirical study. Sojtware Quality Journal 2015; Volume 23(3); pp.393-422. 18. M. Sameh Ibrahim, M. A. R. Mansour, A. M. Ahed. Improve six-sigma management by forecasting production quantity using image_verification reliability tool. International journal of advance in engineering and technology (IJAET) 2011; Volume 1(53); pp332-342, ISSN 2231-1963. 19. Muhamad Safiih Lola and et al. Improving The Performance Of Ann-Arima Models Eor Predicting Water Reliability In The Offshore Area OfKuala Terengganu, Terengganu, Malaysia. Journal of Sustainability Science and Management2018; Volume 13(1); pp 27: 37. 20. J. Sterba and K. Hilovska. The implementation of hybrid ARIMA neural network prediction model for aggregate water consumption prediction. Aplimat—Joumal of Applied Mathe-matics 2010; Volume 3(3); pp.123-131. 21. Reuben Feinman, Ryan R. Curtin, Saurabh Shintre, and Andrew B. Gardner. Detecting adversarial samples from artifacts, 2017. Technical Report. http://arxiv.org/abs/1703.00410. 22. B. Zhu, B. Baesens, A. e. Backiel, and S. K. vandenBroucke. Benchmarking sampling techniques for imbalance learning in churn prediction. Journal ofthe Operational Research Society 2018;Volume 69( 1); pp. 49-65. 23. S. Kaparthi, and N. C Suresh. Machine-component cell formation in Group technology: A neural network approach. International Joumal of Production researchl992;Volume 30(6); pp. 1353 - 1367.[Online], Available: http://www. tandfonline. com/doi/abs/10.1080/0020754920894296l#preview [Sept. 1, 2013] Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11 Artijicial Poka-Yoke Increases Process Reliability Via A Hybrid of the Neural Network with Arima Results 953 24. Qiu, M., & Song, Y. Predicting the Direction of Stock Market Index Movement Using an Optimized Artificial Neural Network Model. PLoS ONE2016; Volume 3(11); pp. 1-11. 25. N. Merh, V. P. Saxena, and K. R. Pardasani. A comparison between hybrid approaches ofANN and ARIMA for Indian stock trend forecasting. Joumal of Business Intelligence 2010; Volume 3(2); pp. 23^3. 26. Xue, T., Peng, Q. F., Huang, G. J., et al. Fault Analysis and Evaluation Method of Ele-vator Door System Based on Fault Tree. Automation & Information Engineering 2015; Volume 6; pp. 34-36. 27. K. Julian, J. Lopez, J. Brush, M. Owen, andM. Kochenderfer. Policy compressionfor air crafl collision avoidance systems. In Proc. 35th Digital Avionics System Conf. (DASC) 2010; pp. 1-10. 28. Uluskan M. A Comprehensive Insight into the Six Sigma DMAIC Toolbox. International Journal of Lean Six Sigma 2016; Volume 7(4); pp. 406-429, doi: 10.1108/IJLSS-10-2015-0040. 29. Uluskan M. Analysis ofLean Six Sigma Toolsfrom a Multidimensional Perspective. Total Reliability Management & Business Excellence2017, doi;10.1080/14783363.2017.1360134. 30. S Ajakaiye, A. B Adeyemo, A. O Osofisan, and O. P A Olowu, O. P. A. Analysis of Poultry Birds production performance usingArtificial Neural Networks. Asian Journal of Information Technology its applications 2006;Volume 5(5).[Online], Avaialable: http://docsdrive.com/pdfs/medwelljournals/ajit/2006/522- 527.pdf [Sept., 12, 2013 31. Verrieir B., Rose B., and Caillaud E. Lean and Green Strategy: the Lean and Green House and Maturity Deployment Model. Joumal of Cleaner Production2016; Volume 116; pp. 150-156, doi:10.1016/j.jclepro.2015.12.022. 32. M. Sridhar 1, K. Padma Raju2, Ch. Srinivasa Rao3, D. Venkata Ratnaml. Prediction and Analysis of Rain Attenuation using ARIMA Model at Low latitude Tropical Station. International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering 2013; Volume2( 7). 33. Huang, S. H. and Zhang, H. -C. Neural-expert hybrid approach for intelligent manufacturing: a survey. Computer in Industry 1995; Volume 26; pp. 107-126. 34. W. Ma, S. Fang, B. Su, X. Xue, M. Li.. (2017). “Second-law-based analysis of vapor-compression refrigeration cycles”: analytical equations for COP and new insights into features of refrigerants Energy Convers Manage 2017; Volume 138; pp. 426-434. R. SherlineJesie. Multi-model Authentication System using Artificial Neural Network. International Conference on Emerging Technology Trends (ICETT), Proceedings by International Journal of Computer Applications 2011, (IJCA). L. Pulina and A. Tacchella. Challenging SMT solvers to verify neural networks. AI Communications 2012; Volume 25(2); pp. 117-135. 37. Z. Huang, H. Yu, Z. Peng, M. Zhao., (2015). “Methods and tools for community energy planning: a review Renew Sustain Energy. ”, Rev 2015; Yolume 42; pp. 1335-1348. 38. Zhang, A. Reliability improvement through Poka-Yoke: from engineering design to information system design. Int. J. Six Sigma and Competitive Advantage2014;Volume 8(2); pp.147-159. 39. Luxhoj, et al., 1996Time series forecasting by combining RBF networks, certainty factors, and the Box-Jenkins model 1996; Volume 10 (2), pp.149-168 • 40. Lazonick, W. and Prencipe, A. Dynamic capabilities and sustained innovation: strategic control andfinancial commitment at Rolls-Royce Plc. Industrial and Corporate Change 2005; Volumrl4; pp. 501-542. 35. 36. www.twrc.ors SCOPUS Indexed Journal editor@tjprc. org 954 Ahmed Abed, Samia Elattar, Tamer Gaafar & Fadwa Alrowais 41. “Reduction of Assembly Time in RAPMAN 3.1 3d Printer by Redesign of One of the Eight Corners by using DFMA Method”, International Journal of Sales & Marketing Management (IJSMM), Vol. 5, Issue 6, pp. 1-6 42. “Inventory Management Practices Followed in Coconut Oil Mills in Western Tamil Nadu”, International Journal of Humanities and Social Sciences (IJHSS), Vol. 5, Issue 2, pp. 23-28 43. “The Integration of Lean Manufacturing and Lean Maintenance to Improve Production Efficiency”, IJMPERD, Vol. 9, Issue 1, pp. 601-612 44. “Role of Lean Manufacturing Leadership on Technology Transfer in India: A Facet of Manufacturing Industries with Reference to Guntur District, Andhra Pradesh”, IJMPERD, Vol. 8, Issue 3, pp. 257-266 Impact Factor (JCC): 8.8746 SCOPUS Indexed Journal NAAS Rating: 3.11