The RSM-CCD approach can predict the optimal polishing conditions to obtain PYKF with the best characteristics.

This study applies the optimum of MMYKF through 40 mesh sieve (13.65% w.b. moisture content) derived from yellow konjac chips milled using micro-mill assisted cyclone separator as the raw materials. Yellow konjac chips were purchased from farmers growing yellow konjac in Klangon Village, Saradan District, Madiun Regency, East Java, Indonesia (Figure 1 and Figure 2).

Potassium permanganate (Sigma-Aldric, USA), D-Glucose (Sigma-Aldric, USA), distilled water, H₂SO₄ (Merck KGaA, Darmstadt, Germany), NaOH (Merck KGaA, Darmstadt, Germany), HCl (Merck KGaA, Darmstadt, Germany), ammonium hydroxide (Merck KGaA, Darmstadt, Germany), methyl red (Sigma-Aldric, USA), CaCl₂ (Merck KgaA, Darmsadt, Germany), and other chemicals were purchased in analytical grade from Merck KGaA, Darmstadt, Germany, and used without further purification.

Color reader (Minolta CR-10, Minolta, Japan), analytical balances (Mettler Toledo Inc., USA), Shaker water-bath (Memmert WNB14, Germany), Centrifuge (Thermo Scientific SL40R, Thermo Fisher Scientific, Inc.), viscometer (NDJ-1 rotational viscometer, China), Shimadzu UV-Visible Spectrophotometer UVmini-1240 (Shimadzu cooperation, Kyoto, Japan), Oven (Memmert, Germany), Vortex (IKA Vortex 3, ProfilLab24 GmbH, Berlin, Germany), Electrical Stove (Maspion, Indonesia), Shaker (SI-1500 Orbital-Genie Orbital Shaker, Marshall Scientific LLC)., Centrifugal mill (Workshop local made, CM 1 model, Indonesia), Disk mill (FFC 45 model, China), Micro-mill assisted cyclone separator (Workshop local made, MM 1 model, Indonesia).

Sample preparation: The study involved the polishing process described by Witoyo et al. (2020) using a centrifugal mill (Figure 2) with minor modifications. The 8.96 – 16.04 kg of MMYKF from the micro-mill, assisted by cyclone separator, was weighed using the digital balance (CMOS digital scale modelDS-30k) and put into input hopper (Figure 2, (4)). The feed rate (20 Hz) and fan blower speed (10 Hz) panel were turned on for 30 seconds for conditioning, and the rotor speed was set to 30 Hz for the polishing process. Each cycle process was completed (30 seconds after CYKF run out in the input hopper). Then the PYKF (the heavy fractions) was collected (Figure 2, in front of (13)) (1^st polishing cycle) and put into the same input hopper for 2 to 8 polishing cycles using the same procedure as in the 1^st polishing cycle based on experimental design (Table 1). The final PYKF (about 250 g) from each process was collected, sampled, and stored in a dark bag for further analysis.

A number of samples analyzed: 13 samples (Table 3) and 3 samples from the verification experiment (Table 2). So there were 16 samples to be analyzed.

A number of repeated analyses: Each sample was in Duplo analyses.

A number of experiment replication: The experiment was conducted using the CCD method, whereby 13 runs for the whole experiment consisted of four replications for the correlation treatment, four replications for the axial treatment and five replications for the centre treatment. The verification experiment used three replications to validate the optimum prediction treatment suggested by Design- Expert program 10 version trials.

Calcium oxalate content in yellow konjac flour was evaluated based on methods described by Iwuoha and Kalu (1995). Briefly, 2 g of yellow konjac flour was mixed with 10 mL of HCl and 190 mL of distilled water, followed by heating at 100 °C for 1 h. The mixed solution was cooled and diluted with distilled water to 250 mL using a volumetric flask. The mixture was filtered and divided into two portions. Each filtrate portion was added with 4 drops of indicator methyl red and ammonium hydroxide (NH₄OH) until the color turned yellow. Then, it was reheated until it reached 90 °C, cooled, and filtered again. Each final filtrate was reheated until it reached 90 °C, and 10 mL of 5% CaCl₂ was added. The mixed solution was stored at 5 °C overnight. Each filtrate was centrifuged at 8000 rpm for 15 min to obtain pellets. The pellets were dissolved in 10 mL of 20% H₂SO₄ to obtain 10 mL of filtrate from each filtrate. The two filtrates were mixed and diluted with distilled water to 250 mL using a volumetric flask. 125 mL of mixed filtrate solution was then heated to almost boiling and titrated with 0.05 M standardized KMnO₄ to form a pink color and persisted for 30 seconds. The calcium oxalate content in yellow konjac flour was calculated by Eq. 1.

(1) C a l c i u m   O x a l a t e   C o n t e n t ( % ) = T × ( V m e ) ( D f ) × 10 5 ( M E ) × M f × 1000

Where: T is the volume of KMnO₄ used for titration (mL), V_me is the equivalent mass volume (1 cm³ 0.05 M KMnO₄ equivalent to 0.0025 g anhydrous oxalic acid), Df is a dilution factor, ME is the molar equivalent of KMnO₄ solution (0.05), and Mf is the mass of the yellow konjac flour (g).

A solution (1% w/v) of yellow konjac flour was mixed using a constant temperature water-bath (75 °C) (Memmert, Germany) with a constant agitation speed (150 rpm) until perfectly hydrated (Liu et al., 2002). The measurements were performed at room temperature using an NDJ-1 rotational viscometer with 12 rpm stirring speed and the spindle number 4 (Witoyo et al., 2020).

Yellow konjac flour color was evaluated using Minolta CR-10 (Minolta, Japan). The yellow konjac flour's surface was monitored and expressed as Hunter value (L, a, and b). The DoW was calculated based on measured Hunter value using Eq. 2 (Witoyo, Widjanarko and Argo, 2019).

(2) Degree of Whiteness=100 - ( 100-L ) 2   +   a 2   +   b 2

With D-glucose as a standard curve, the spectrophotometric method was used to determine the glucomannan content in the yellow konjac flour (Chua et al., 2012). In brief, 200 mg of yellow konjac flour sample was dissolved in 50 mL of NaOH-formic acid buffer, and mixed overnight at room temperature. The mixture was dissolved using a NaOH-formic acid buffer to 100 mL using a volumetric flask, and followed by centrifugation for 4000 rpm for 15 minutes to obtain glucomannan extract. The glucomannan hydrolyzate was obtained by hydrolyzing 5 mL of glucomannan extract with sulfuric acid (2.5 mL, 3 M), followed by boiling for 1.5 h in a water-bath. The mixture was cooled and added with NaOH solution (2.5 mL, 6M). The mixture was diluted again to 25 mL in a volumetric flask using distilled water. The absorbance of both the glucomannan extract and hydrolyzate was measured using a spectrophotometer at 540 nm. Compared with the D-glucose standard curve to obtained the glucose content (mg) in both samples. The calculation of glucomannan content (% d.b.) was following Eq. 3.

(3) GM content (% d. b. ) = 0.9× ( 5T-T 0 ) ×f m× ( 100-w ) ×100

Where: m is the mass of yellow konjac flour (mg), w is water content in yellow konjac flour (%), T_o is glucose content in yellow konjac flour extract (mg), T is glucose content in yellow konjac flour hydrolysate (mg), and f is dilution factor.

All physicochemical properties of 13 samples of polished yellow konjac flour (calcium oxalate content, viscosity, DoW, and glucomannan content) were analyzed in duplicate, as shown in Table 3.

The CCD-RSM was applied in the polishing process of MMYKF using the centrifugal mill. At the same time, the MMYKF mass and polishing cycle were the independent variables. Following the report by Witoyo et al. (2020), this study's central point was 12.5 kg with five times for MMYKF mass and polishing cycle, respectively. The range of independent variables used in this study is shown in Table 1. Physicochemical data of 13 randomized polishing processes were analyzed using Design-Expert 10 trial versions (State Ease, Inc.) based on multivariate analysis. The quadratic equation was used to calculate non-linear models of the observed system (Faridah and Widjanarko, 2013), as shown in Eq 4. The appropriate model was determined based on R², adjusted R², lack of fit, and p-value of the model from the Design-Expert program's calculation (Sugiono et al., 2019). The verification (Table 6) was carried out with three replications under the optimum polishing conditions prediction from the laboratory's Design-Expert program. The software predictions and verification data were analyzed and compared by paired T-test using Minitab 17 (Stat View, USA) to determine the optimum polishing conditions' accuracy.

(4) Y = β 0 + ∑ i = 1 2 β i X i + ∑ i = 1 2 β i i X i 2 + ∑ i = 1 , j = 2 k − 1 , k β i , j X i X j + ε

Where ε is a random error, k is the number of independent parameters, X_i and X_J are the coded independent variables of MMYKF mass and polishing cycle, β₀, β_i, β_ii, and β_ij are the regression coefficients of the intercept, linearity, quadratic, and interaction of the treatment, respectively, and Y is the observed response.

The polishing time of each run of polishing is shown in Table 2. Table 2 shows that the MMYKF mass and number of polishing cycles increased the time of polishing. The longest polishing time was found at MMYKF mass of 15 kg and a polishing cycle of seven times with 59.55 minutes, whereas the shorter polishing time was found at MMYKF mass of 12.5 kg and the polishing cycle of two times with 16.25 minutes.

In this study, MMYKF mass (X₁) and polishing cycle (X₂) were applied to optimizing four responses of PYKF. The four PYKF responses were calcium oxalate, viscosity, DoW, and glucomannan content (Table 3). The selection model was evaluated for each response by the R², adjusted R², lack of fit, and p-value. Considering the criteria, the quadratic model was suitable and adequate to represents data from all responses of PYKF with a p-value <0.05. Table 4 shows the fitting model, coefficient regression, and significance of the second polynomial order for all responses. The R² of quadratic models were 0.9680, 0.9647, 0.9469, and 0.9808 for calcium oxalate, viscosity, degree of whiteness, and glucomannan content, respectively. Moreover, the adjusted R² value for calcium oxalate, viscosity, DoW, and glucomannan content in the sequence were 0.9452, 0.9394, 0.9090, and 0.9672. The higher R² and adjusted-R² indicated the quadratic model's adequacy to explain all response data (Mohapatra and Bal, 2010). Moreover, for all responses the lack of fit (p ³0.05) values were calculated, indicating that the quadratic model was adequate to predict all observed responses ( Faridah, 2016; Mohapatra and Bal, 2010). The lack of fit for calcium oxalate, viscosity, DoW, and glucomannan responses were 0.7053, 0.3746, 0.2652, and 0.0521, respectively.

The Effect of Independent Variables on Physicochemical Properties of PYKF

Calcium oxalate is one of the non-glucomannan components that is surrounding glucomannan cell granules and is toxic when consumed; thus, it needs to be removed. Calcium oxalate content in PYKF ranged from 0.45 to 1.12 % w.b. The present study results were higher than the data reported by Faridah (2016) and lower than the research data by Witoyo, Widjanarko and Argo (2019). In this study, calcium oxalate was affected merely by the number of polishing cycles, as illustrated in Figure 3A. Calcium oxalate content decreased quadratically as a function of the polishing cycle increased. The decrease of calcium oxalate might be caused by erosion of non-glucomannan compounds covering the surface of yellow konjac flour granule during the polishing process, due to the frictional force between the rotor on the centrifugal mill and yellow konjac flour, and between yellow konjac flour, followed by the continuous fractionation process during the process (Witoyo et al., 2020). According to Nurhasanah, Antarlina and Syah (2017) and Nurhasanah et al. (2019), the polishing process can reduce the anti-nutritional compounds (tannins) in whole sorghum grains. Wachyuningsih (2011) stated that polishing yellow konjac flour using an abrasive polisher reduces the calcium oxalate in yellow konjac flour. Moreover, the calcium oxalate of PYKF was not affected by the MMYKF mass.

Viscosity is one of the important indicators in the utilization of yellow konjac flour in various applications, particularly in food products, as thickening agents, gelling agents, stabilizers, emulsifiers (Impaprasert, Borompichaichartkul and Srzednicki, 2014), and fat replacers (Jiménez-Colmenero et al., 2013), which require high viscosity. The viscosity of PYKF was in the range of 11.125 – 20.750 cP. This data was higher than the data reported by Faridah (2016) and lower than the data reported by Xu et al. (2014). The data finding indicated the PYKF included in the second the first-grade classification for common konjac flour, which the Ministry of Agriculture issued, People's Republic of China (Liu et al., 2002). The polishing cycle performed a quadratic effect on PYKF viscosity, as illustrated in Figure 3B. The PYKF viscosity increased five times in the polishing cycle and decreased afterward.

In the study, glucomannan content affected the PYKF's viscosity. The higher the glucomannan content in PYKF the higher viscosity of its. Based on Faridah (2016) and Wardhani et al. (2016), viscosity is positively correlated with glucomannan enclosed in yellow konjac flour. Ryan, Yuan and Crosby (2004) added that glucomannan had a higher viscosity than other polysaccharides at the same concentration. Besides, the MMYKF mass had not affected the viscosity of PYKF.

The DoW of PYKF was in the range of 59.94 – 62.38. This result was lower than the results reported by Impaprasert, Borompichaichartkul and Srzednicki (2014) and higher than data reported by Witoyo, Widjanarko and Argo (2019) and Witoyo et al. (2020). The DoW of PYKF was merely affected by the polishing cycle, as illustrated in Figure 3C. The polishing cycle provided a positive quadratic trend to PYKF's DoW. Decreasing impurity (non-glucomannan) components covering yellow konjac flour granules during the polishing process increase PYKF DoW. This result was consistent with the previous study by Nurhasanah and her team who claimed that the polishing process can increase the DoW of sorghum using abrasive polishing machines or modification of polishing machines with three-cylinder levels (Nurhasanah, Antarlina and Syah, 2017; Nurhasanah et al., 2019). Reddy et al. (2017), and Paiva et al. (2016) noted that improving color or DoW was closely related to the reduction of certain impurities during the milling process. Witoyo et al. (2020) reported that that the content of non-glucomannan compounds, such as proteins, fat, calcium oxalate, and starch had decreased during the polishing process of yellow konjac flour. However, in the present study, the MMYKF mass had a not significant effect on PYKF's DoW.

Glucomannan is the most important parameter to determine the quality of yellow konjac or konjac flour based on international standards. According to the international standard for konjac flour, particularly concerning common konjac flour, glucomannan is divided into three classes, namely top-grade, first-grade, and second-grade with glucomannan content ³75%, ³65%, and ³60% d.b, respectively (Liu et al., 2002). Based on these standards, PYKF is ranking second in the category of first-grade common konjac flour with glucomannan content in the range of 60.72 – 69.92% d.b. This result is lower than KGM content by wet extraction from yellow konjac flour, as reported by Impaprasert, Borompichaichartkul and Srzednicki (2013); Wardhani et al. (2016) and similar to Faridah, Widjanarko and Sutrisno (2011) study.

In the present study, the polishing cycle affected glucomannan content. The number of polishing cycles exerted a quadratic effect on the glucomannan response, as shown in Figure 3D. The increase of the number of polishing cycles to five enhanced the glucomannan content, then decreased subsequently. The increased number of polishing cycles reduces distinct compounds, such as calcium oxalate, starch, protein, and other impurities surrounding the glucomannan cells in yellow konjac flour, due to friction between rotors of the centrifugal mill and flour as well as between the flour particles. According to Witoyo et al. (2020), the polishing process destroys non-glucomannan compounds on the glucomannan cell surface in yellow konjac flour. Subsequently, the nonglucomannan components were detached and separated from glucomannan granules using a cyclone separator. Moreover, glucomannan content decreased after polishing treatment under the optimum conditions due to attrition of larger glucomannan particles that turned into fine particles and were eliminated in the cyclone separator. This finding was supported by Faridah, Widjanarko and Sutrisno (2011) and Wachyuningsih (2011) after milling yellow konjac chips using the stamp mill, and polishing yellow konjac flour using an abrasive polisher, respectively. Furthermore, the MMYKF mass had not affected the glucomannan of PYKF.

Polishing parameters were optimized to produce PYKF (Figure 1C) with high viscosity, DoW, glucomannan content, and low calcium oxalate content by using the centrifugal mill. The independent variables and response criteria to achieve the desired characteristics of PYKF are compiled in Table 5. The predicted optimum polishing condition was obtained at 10.00 kg and 6.17 polishing cycles (Table 6), with the following characteristics: 0.52 ±0.00% w.b. calcium oxalate, 20362.00 ±16.00 cP viscosity, 62.22 ±0.01 DoW, and 69.43 ±0.02% d.b. glucomannan content with desirability value of 0.927 (Figure 3E). The verification with modified conditions (Table 6) obtained the following characteristics: 0.56 ±0.05 % w.b. calcium oxalate, 20208.00 ±564.00 cP viscosity, 62.16 ±0.29 DoW, and 68.18 ±1.28% d.b. glucomannan content. Consequently, the verification data revealed no significant differences (p-value >0.05) for all responses compared to predicted values. This shows that the quadratic model was accurate and appropriate for optimizing yellow konjac flour polishing using the centrifugal mill (Mohapatra and Bal, 2010). The desirability value of 0.927 expressed the prediction accuracy of the optimum polishing conditions equivalent to 92.70%.