Development of eDART-based weight prediction system in injection molding via Taguchi design and fuzzy logic

This research describes developing a Fuzzy Logic based weight prediction system (FL-eWPS) during the process of injection molding. The main purpose is to apply Fuzzy Logic to predict defects during injection molding operations while processing parameters, such as shot size, barrel temperature, cooling time, and holding pressure. The parameters are varied within a shorter range when using Delrin 511 DP plastic from DuPont Engineering Polymers. eDART data logging system was used for real-time data collection for the different parameters by using the sensors during the injection filling stages. A Fuzzy Logic reasoning algorithm was applied to gain the threshold values of weight prediction with various processing parameter settings. During the injection molding process, the FL-eWPS system was shown to predict weight with 99% accuracy.


Introduction
To achieve the mass manufacture of plastic products especially the components with complex geometries that require high quality standards, injection molding process will be considered.On top of that, quality control is critical criterion in achieving high-quality specifications.There are some parameters that have major influence on the process that should be taken into consideration in the injection molding process, such as injection temperature, resin melt properties, etc [1,2].Regarding the previous research, it noted that the disk weight serves as a good quality index for injection molding process [3]: the higher the weight, the stronger the molded disk as air bubbles.Therefore, the injection molding parameters setting affect the product's quality, which the index of weight could indicate the density of the product.Furthermore, the melting temperature and injection pressure have a significant impact on the quality of the plastic disk as well.Based on the previous review, it has been stated that the increasing injection pressure will lead to heavier product weight due to its increased density under high pressure [4].The consistency of injection weight precision is determined by the quality of the mold processing, monitoring parameters, raw material performance, and controlled parameters [5,6].Different statistical methods could be developed by monitoring the stability of injection molding quality over time.The injection molding defective parts are predicted using various online or in-process control systems.Gheorghe et al. [7] In addition, Wang et al. [8], conducted a numerical experiment, and the confirmation run pointed out that the error of yield was less than 4 %.Zhang [9] proposed a statistical quality monitoring method for extracting variables in real time, highlighting the importance and necessity of variable extraction, despite the lack of quantitative relationships among variables.The study of injection weight affects process parameters and the repeatability of weight algorithms, which helps to improve the accuracy of weight consistency [10].The fundamental research included the interested process variables, sample data collection, product quality, and using suitable mathematical approximation algorithms to approximate the relationship.The approaches that are commonly used in linking hidden relationships between aspects are Artificial Neural Networks (ANNs), Partial Least Squares Regression (PLS), Fuzzy logic (FL), Support Vector Regression (SVR), and Gaussian Process (GP).Compared to other algorithms [11], The fuzzy logic algorithm is capable of capturing apprehension or ambiguity related cognitive processes exhibited by humans, such as reasoning or decision-making.Zadeh [12] introduced fuzzy logic in 1965, and since then it has become a well-known theory that drives traditional logic to an advanced level.The traditional logic theory only allows binary sets of inputs: true or false, whereas fuzzy logic theory allows intermediate values whereas fuzzy logic theory allows for intermediate values between standard assessments such as yes/no, black/white, and so on.[13].According to Li and Elbestawi [14], a nonlinear process inherent uncertainty was dealt with in a systematic way using fuzzy logic, which makes it a preferred algorithm.Due to the features of user convenience, easy-to-maintain, fuzzy logic has been utilized in industry for controlling complex systems.Chiang Chang [15] applied a fuzzy logic method to the thin shell injection molded part optimal process design.Aydin and Mehmet [16] developed a controller using fuzzy logic rules for predicting flow length for engineering plastics.Mok and Kwong [17] rely on artificial neural network and fuzzy logic to design the injection molding process.Based on the conclusion of previous research, it pointed out that fuzzy logic predicts reliable results by setting different parameters relatively.Also, the prediction models can distinguish between within-tolerance and out of tolerance parts based on machine and process parameters.

Methodology
To enable predicting the weight of the plastic disk, this research built a prediction model by applying the fuzzy logic inference system.The approach of Taguchi experimental design was used to determine the optima machine parameter setting combination by taking advantage of using an L9 orthogonal array, which is able to screen out the input has the most impact on the outcome.To confront the existing variations in the process, the factorial design of experiment (DOE) was applied to this research; in addition, the optimized parameter setting was based on the conclusion of DOE and was set by the expanded parameter scale.

Experimental design
The Engel injection molding machine hardware system is shown in Figure 2.After the completion of the experimental set-up, the FL-eWP system was ready for evaluation.Taguchi experimental design was applied to this research for improving injection molding process parameters.A L9 orthogonal array was designed with selected parameters being shot size, barrel temperature, cooling time, and holding pressure with three levels each as shown in Table 1.The raw plastic material in this research is Delrin 511DP (Polyoxymethylene), purchased from DuPont Engineering Polymers.The product of this research is shown in Figure 3, and five repetitions for each run.L9 Orthogonal array with designed parameters and the measured weights of the five repetitions for each run showed in Table 3.The Y-bar displays the average diameter of each trail, and S/N ratio was obtained by applying the Formula (1).

𝜂 10𝑙𝑜𝑔
where ƞ is the S/N ratio, y ̅ is the average of weight, and  is the variance.Table 3 demonstrates response table for the diameter and S/N ratio respectively, The setting with diameter closest to the nominal value (8g) and the largest S/N ratio value were chosen for the best parameter setting due to the nature that the larger the ratio the less effect on output.

Fuzzy logic inference system
Gong and Chen proposed a five-step fuzzy logic inference system by using MATLAB fuzzy logic toolbox [18]: Firstly, to import the data frame with the input variables, average mold temperature, fill and pack integral end of cavity pressure, processing time, and cavity fill end of cavity, and the output variable, weight Secondly, to determine the membership of the inputs and outputs.It is determined as a curve function which correlates each input point to the value sets were developed by combining input and output values in linguistic terms and membership function grades.MATLAB Fuzzy Logic toolbox [19] has eleven membership function types built in.Thirdly, to develop the fuzzy rule bank.The rule bank from the fuzzy logic inference system is built based on the domain knowledge or the empirical training data [13].Fuzzy rules collect lingual statements that record Fuzzy logic inference system behavior, whether to classify an input, or control the output.The form of Fuzzy rules are as follows (2): if (membership function1(input1)) and/or (membership function2(input2)) and/or?then (output membership function (output)). ( Then, to build the logic of decision-making.Certain types of IF-THEN rules will be triggered to give feedback based on the condition.The response of compensation for each rule will be given according to the degree of membership.Finally, to complete the process of Defuzzification.Due to the output of the fuzzy logic inference element in the table of non-numerical value that is not readable for machine controller; the fuzzy value is interpreted into sharp value through defuzzification process.The prediction model performance will be verified by validation test though fuzzy inference system has been built.The testing data is used to predict the results of weight.By comparing the results between the true weight of experiment and the predicted weight from the model, it can be concluded that the performance of the obtained prediction model.

Development of FL-EWP system
In this secession, Regression analysis is used to evaluate the weight of output parameters.From this standpoint, it could be noted that three input variables, Mold temperature (MT), EOC pressure, and EOC Time, show significant impact on the output variable, weight (W).The ranges for input and output variables were as follows: The triangular curve was selected for both input and output membership functions for determination, shown in Figure 4 and Figure 5. Once the membership function gets determined in the Fuzzy logic toolbox, the Fussy rule bank will be defined.The inference element used IF-THEN logic to determine the output weight.Regardless of the noise factor's potential effects on the response variable, to determine the optimal parameter, the highest value in the signal to noise ratio response table was chosen.An example of If-Then rule 27, in The Fuzzy Logic inference system has been validated by three input variables, mold temperature, EOC Pressure, and EOC temp, concluding that the performance of the prediction model has a good fit with the actual data in 99% of accuracy.

Conclusion
In this research, the parameters in the injection molding process were analyzed and optimized based on the Taguchi approach and design of experiment.According to the results of the experiment, the prediction system for weight was developed by applying the Fuzzy Logic toolbox-based system.Moreover, based on the comparison of predicted model and actual data, it can be concluded that the predicted outcomes have a congruency with the actual outcomes, showing that the prediction model could successfully predict weight with over 99% accuracy by inputting metering stroke (shot size), barrel temperature, holding pressure, and cooling time for a small range variation.

Figure 1 .
Figure 1.Fl-EWP system structure model.The eDART data logging system collected the real-time data of each injection molding process for developing the Fuzzy logic-based prediction system.Then, the value predicted by the system was compared with the real value obtained by the optimized parameter setting, evaluating the performance of the predictive model.

Table 1 .
Taguchi parameter design with 4 factor and 3 levels each.

Table 2 .
L 9 Orthogonal array with Taguchi parameters.

Table 3 .
Taguchi experimental data and analysis.

Table 4 .
Response table for signal to noise ratio.

Table 5 .
Response table for diameter of the disc.

Table 6
If have Mold Temp is high and if input Fill & Pack Integral, End of Cavity EOC Pressure is high, and if Process Time, Cavity Fill EOC is high, then output weight should be Out of Tolerance on the upper side with the weight 1."According to the rule viewer the MATLAB Fuzzy logic toolbox, Figure 6, there are one hundred and eight rows with four separated columns that show Mold Temperature, EOC Pressure, EOC Time, and the weight respectively, following the rule, that IF the part is represented by the first three columns and THEN the part is represented by the last column.Ave Mold Temp, MT (•C) → [140, 160] Fill & Pack Integral, End of Cavity EOC Pressure, EOCP (BAR) → [800, 840] Process Time, Cavity Fill End of Cavity EOC, EOCT (S)→ [0.514, 0.524] W (g) →[ 7.80, 8.20]