Product Quality Output Measurement for Preventive Maintenance on Computer Numerical Control (CNC) Machines at an Electronic Manufacturing Industry Aizat Haikal Bin Apandi1*.
Adam Shariff1.
Kamarulzaman Mahmad Khairai2 Centre Science and Mathematics.
University Malaysia Pahang As-Sultan Abdullah.
Pahang.
Malaysia Kuliyyah of Engineering.
International Islamic University Malaysia.
Selangor.
Malaysia * Correspondence: azthkl05@gmail.
ABSTRACT
Received: 8 October 2024 Revised: 19 December 2024 Accepted: 30 December 2024 Citation:
Apandi.
Shariff.
, &
Khairai.
Product Quality output measurement for preventive maintenance on computer numerical control (CNC) machines at an electronic manufacturing industry.
QOMARUNA Journal of Multidisciplinary Studies, 2.
, 1Ae19.
Computer Numerical Control (CNC) machines remove material from a blank or workpiece using digital controls to produce custom-designed parts.
Maintaining their accuracy and precision under challenging conditions after long-term usage is crucial.
This study aims to evaluate CNC product quality using Overall Equipment Effectiveness (OEE) and enhance long-term performance through data-driven approaches.
The method of this study focuses on analyzing scrap rate data, employing a u-chart to monitor stability, and applying machine learning regression modelsAiK-Nearest Neighbour (KNN) and Random Forest (RF)Aito forecast scrap rates.
These forecasts help identify when preventive maintenance is necessary, preserving machine precision over time.
This study also applied visualization of results with Microsoft Power BI to enhance data interpretation, aiding quick responses to potential problems.
Results indicate that RF outperforms KNN in predicting scrap rates.
Stacking these models further improves accuracy, offering a more reliable decision-making tool for anticipating quality issues.
By detecting anomalies early, manufacturers can implement timely maintenance, minimizing downtime and prolonging CNC machine In conclusion, integrating scrap rate analysis, statistical process control, and advanced machine learning techniques can maintain product quality and reduce Companies should include more proactive maintenance planning by employing better forecasting.
Keywords: Computer Numerical Control (CNC).
Overall Equipment Effectiveness (OEE), preventive maintenance, quality improvement, machine learning.
Copyright: A 2024 by the Submitted for possible open-access publication under the terms and conditions of the Creative Commons Attribution (CC BY) license .
ttps://creativecommons.
org/lic enses/by-nc-sa/4.
0/).
ABSTRAK
Mesin Computer Numerical Control (CNC) menghilangkan material dari sebuah benda kerja dengan menggunakan kontrol digital untuk menghasilkan komponen yang dirancang secara khusus.
Penting untuk menjaga akurasi dan presisi mesin ini di bawah kondisi yang menantang setelah penggunaan jangka panjang.
Penelitian ini bertujuan untuk mengevaluasi kualitas produk CNC menggunakan Overall Equipment Effectiveness (OEE) dan meningkatkan kinerja jangka panjang melalui pendekatan berbasis data.
Metode penelitian ini berfokus pada analisis laju scrap, penggunaan u-chart untuk memantau stabilitas, serta penerapan model regresi machine learningAiK-Nearest Neighbour (KNN) dan Random Forest (RF)Ai untuk memprediksi laju scrap.
Prediksi tersebut membantu mengidentifikasi waktu yang tepat untuk melakukan pemeliharaan pencegahan, sehingga presisi mesin dapat tetap terjaga seiring waktu.
Penelitian ini juga memanfaatkan visualisasi hasil menggunakan Microsoft Power BI untuk meningkatkan interpretasi data dan memfasilitasi respons cepat terhadap potensi masalah.
Hasil penelitian menunjukkan bahwa RF memiliki kinerja lebih baik dibandingkan KNN QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
dalam memprediksi laju scrap.
Penggunaan stacking pada model-model tersebut lebih lanjut meningkatkan akurasi, sehingga memberikan alat pengambilan keputusan yang lebih andal dalam mengantisipasi masalah kualitas.
Dengan mendeteksi anomali secara dini, produsen dapat melakukan pemeliharaan tepat waktu, meminimalkan waktu henti, serta memperpanjang umur operasional mesin CNC.
Kesimpulannya, integrasi analisis laju scrap, pengendalian proses statistik, dan teknik pembelajaran mesin yang canggih dapat secara efektif menjaga kualitas produk serta mengurangi ketidakakuratan.
Perusahaan sebaiknya mengadopsi perencanaan pemeliharaan yang lebih proaktif dengan memanfaatkan peramalan yang lebih baik.
Keywords: Computer Numerical Control (CNC).
Overall Equipment Effectiveness (OEE), preventive maintenance, perbaikan kualitas, machine learning Introduction Nowadays, manufacturing companies supplying electronic components recognize the critical importance of product quality to maintain customer satisfaction and competitiveness.
Ensuring product quality begins with maintaining precise, reliable production equipment.
Preventive maintenance plays a key role in improving machines' Overall Equipment Effectiveness (OEE), thereby enhancing performance, availability, and product quality.
Moghaddam .
defines preventive maintenance as scheduled activities over a planning horizon to extend a systemAos useful life and maintain responsiveness, ultimately making it more reliable and available.
Within manufacturing, computer numerical control (CNC) machines form the backbone of precision-driven production processes.
CNC technology removes material layers from a workpiece using digitally controlled machine tools, enabling high-accuracy, custom-designed parts.
Over extended operation, maintaining precision under challenging conditions becomes essential to ensure product quality (Liu, et al, 2.
Preventive maintenance strategies for CNC machines help preserve this accuracy and reduce product scrap.
This benefits both the manufacturerAos profitability and The transformative potential of predictive maintenance has been extensively demonstrated in prior research.
Shahin et al.
investigated over 20 fault detection models using Machine Learning (ML).
Deep Learning (DL), and Deep Hybrid Learning (DHL) to minimize manufacturing downtime, achieving significant accuracy improvements in early failure detection.
Traini .
introduced a machine learning framework for predictive maintenance in milling, leveraging Industrial IoT and AI technologies to enhance real-time monitoring and defect detection.
Similarly.
Paolanti .
employed Random Forest models within a Machine Learning architecture to improve system reliability and prevent unexpected equipment failures.
These studies emphasize the effectiveness of advanced analytics in optimizing maintenance practices, reducing defects, and improving reliability.
Despite these advancements, case-specific investigations remain critical, particularly for established companies like AuCompany T,Ay an electronic manufacturer located in Pahang.
Malaysia, serving the aerospace, automotive, medical, and industrial sectors.
Although the company has operated for over two decades, it has yet to implement predictive maintenance practices.
Addressing this gap offers an opportunity to tailor state-of-the-art methodologies to real-world challenges, ensuring consistent product quality and operational efficiency.
This study aims to investigate the application of preventive maintenance and predictive tools in improving CNC machine quality control, particularly for air coils, which are critical components for automotive and industrial clients.
By analyzing measurement data, employing advanced statistical and machine learning approaches, and implementing systematic maintenance strategies, this research aims to minimize product scrap, enhance OEE, and improve customer satisfaction while providing practical solutions tailored to the specific operational conditions of AuCompany T.
Ay QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
Literature Review Computer Numerical Control (CNC) Machine Computer Numerical Control (CNC) machines reflect a regionAos industrial development and are central to advancing aerospace, military, automotive, and electronic manufacturing.
They have long played a key role in national competitiveness.
Among their performance measures, machine accuracy is paramount, as a partAos precision ultimately depends on the machineAos machining accuracy (Liu et al.
Their accuracy, which relates to how closely finished parts match design specifications, is a key performance measure.
As noted in "Monitoring of CNC Machine Tool Accuracy," proper maintenance and adherence to certified standards (ISO.
DIN.
EN) ensure machines meet precision requirements.
Perfect accuracy is rarely attainable.
HolubAos "Geometric Error Compensation of CNC Machine Tool" highlights that actual part dimensions seldom match nominal design values due to process Finishing operations, however, can bring measurements closer to the target.
Manufacturers enhance product precision and maintain high-quality standards by emphasizing quality measures and reducing output errors.
Figure 1.
Computer Numerical Control (CNC) machine (Documentation of Company AuTA.
Overall Equipment Effectiveness (OEE) Overall Equipment Effectiveness (OEE) is an excellent indicator for measuring sustainability improvements relative to a companyAos initial operational state.
As a key performance indicator (KPI).
OEE reflects not only a machine or systemAos performance, but also the effectiveness of the personnel responsible for its maintenance (Haddad et al.
, 2.
It integrates aspects of operation, maintenance, and resource management (Tsarouhas, 2.
Although OEE could not reveal the exact causes of inefficiencies, it helps categorize areas needing improvement (Tsarouhas, 2.
Enhancing OEE can boost production capacity, improve product quality, reduce downtime, and increase overall system efficiency.
According to P.
Tsarouhas .
OEE is determined by three critical parameters: availability, performance, and quality.
Figure 2 illustrates how OEE is calculated based on these factors.
Measuring OEE effectively evaluates the efficiency of a single machine or an integrated manufacturing system (Dewi, et al.
, 2.
Widely recognized as a measure of internal efficiency.
OEE reflects a machineAos true value-added output.
It helps identify equipment-related losses to improve overall asset performance and reliability.
These losses can be grouped into six major categories (Tsarouhas, 2.
: equipment failure, setup and adjustment delays, minor stoppages and idle times, speed reductions, defects or rework, and reduced performance (Tsarouhas, 2.
By addressing these losses, improving OEE results in fewer breakdowns, reduced idle times, lower defect rates, and fewer workplace accidents.
It also boosts productivity, optimizes processes, encourages workforce involvement, increases profits through cost QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
savings, enhances customer satisfaction, and raises employee morale and confidence.
Figure 3 illustrates these six major categories.
Figure 2.
OEE Calculation (Tsarouhas, 2.
Figure 3.
Six Major OEE (Tsarouhas, 2.
Quality Quality is one of the three factors used to determine OEE.
The quality rate measures how many non-defective products a machine produces during its run time (Dewi, et al.
, 2.
It evaluates the percentage of satisfactory units produced, accounting for defects, scrap, and rework that affect final product quality.
To calculate quality, count the number of units meeting required standards and divide by the total number produced while operating (Tsarouhas, 2.
Quality= Good Units y 100% Total Units .
Improving quality involves implementing quality control checks, enhancing equipment reliability, and providing better training and procedures.
These measures raise the OEE score by reducing scrap and rework, increasing customer satisfaction, and improving overall product quality.
Performance Performance is one of the three OEE components, measuring how effectively equipment operates relative to its maximum potential output (Singh, et al.
, 2.
It accounts for speed losses, minor stops, and slow cycles that reduce the rate at which products are produced, even if the equipment is fully QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
Performance is calculated by comparing the total products produced to what could have been produced under optimal conditions (Dewi, et al.
, 2.
Performance = Actual Output y 100% Maximum Possible Output Increasing performance can involve maximizing production rates, minimizing downtime between cycles, streamlining changeover times, and providing better training and procedures.
enhancing performance, you can increase both the equipmentAos productivity and its OEE score.
Availability Availability is the third component of OEE and measures the percentage of time equipment is ready for use (Haddad, et al, 2.
It accounts for both scheduled and unscheduled downtime, as well as any other factors that prevent the equipment from operating at full capacity.
Scheduled downtime includes planned maintenance, changeovers, and other predefined events requiring the equipment to be offline.
Unplanned downtime covers unexpected breakdowns, repairs, and incidents that halt normal To determine availability, you must identify the total time the equipment ran productively and compare it to the planned production time during the measurement period.
The formula for availability is:
Availability = Operating Time - Downtime y 100% Operating Time Operational time represents the total period the equipment is intended to run during production, while downtime is the total time it remains unavailable due to unexpected issues like breakdowns or Improving availability involves preventive maintenance, reducing changeover time, enhancing equipment reliability, and implementing effective planning and scheduling.
By increasing equipment availability, you boost overall productivity and the OEE score.
Preventive Maintenance Preventive maintenance refers to scheduled activities performed at regular intervals to extend a systemAos useful life and keep it both productive and responsive (Moghaddam, 2.
These activitiesAi such as inspections, cleaning, lubrication, adjustments, alignments, and component replacementsAi reduce a systemAos Aueffective ageAy and lower failure rates.
Manufacturers typically provide recommended maintenance schedules to minimize unexpected failures over a machineAos operational While preventive maintenance improves reliability and availability, its designers must balance the costs of upkeep and replacement against the risks and expenses of unanticipated breakdowns.
Applied correctly, preventive maintenance ensures CNC machines remain efficient, producing parts that meet precise design specifications and maintaining high quality standards.
By proactively addressing wear and tear, companies can reduce product defects, lower scrap rates, and improve their overall profitability.
Systematic preventive maintenance planning also helps coordinate service intervals, allowing organizations to address issues before they arise and maintain optimal equipment This strategic approach ultimately leads to more consistent product quality, increased customer satisfaction, and better long-term returns.
Predictive Maintenance Predictive maintenance represents a proactive strategy that relies on data analysis and predictive modelling to anticipate equipment failures, enabling maintenance tasks to align with actual conditions rather than fixed schedules.
Accordingly, organizations can reduce downtime and costs, a key driver for its growing adoption across multiple industries, including CNC machine facilities.
QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
In a prior study.
Lee et al.
developed a predictive maintenance system for CNC machines that used Machine Learning (ML) models to forecast cutting tool loads.
By anticipating periods of high stress, maintenance activities could be scheduled more effectively, minimizing downtime and extending tool life.
However, the systemAos success depended heavily on accurate, high-quality sensor data and machine logs.
Expanding on this research.
Soori et al.
reviewed ML and AI techniquesAisuch as regression models, support vector machines (SVM.
, and artificial neural networks (ANN.
Aiapplied to CNC predictive maintenance.
They identified vast potential for ML and AI to revolutionize maintenance practices but noted challenges like machine performance variability and non-standardized The authors called for collaboration between data scientists and domain experts to address these issues and fully exploit ML and AI capabilities.
In another advancement.
Ruiz Rodryguez et al.
introduced multi-agent deep reinforcement learning (MADRL) for predictive maintenance in parallel machines.
This adaptive system outperformed traditional methods by reducing downtime and extending machine lifespan.
Nevertheless, its heavy computational demands may limit its applicability in certain environments.
Method This study involves two main stages: first, constructing a control chart based on defects per unit produced by CNC machines.
second, developing predictive models using various machine learning Control Chart A u-chart, a key tool in statistical process control (SPC), monitors the number of defects per unit over time.
It is particularly useful in quality control and manufacturing, where minimizing defects is essential for cost-effectiveness, customer satisfaction, and compliance with industry standards (Tang, et , 2.
To construct a u-chart, samples are collected at regular intervals, and the defect rate per unit is calculated for each sample.
Care should be taken to ensure that the sample size .
is representative, as variations in n affect the accuracy of the results.
In cases where sample sizes differ, applying a weighted moving average can help maintain a consistent view of process performance.
Equation .
calculates the average defect rate per unit, which serves as the central line in the u-chart whereas Equation .
provides the formulas for the Upper and Lower Control Limits (UCL and LCL).
UCL and LCL reflect the acceptable range of variation for an in-control process.
If data points fall outside these limits, it may indicate an out-of-control condition, prompting further investigation (Ieren, et al.
, 2.
yc ycu Ocyca yc= Ocycu yc= yuaycy = .
yc ycu c: number of defects k: number of lots n: sample size Upper Control Limit:
ycOyaya = yc 3yuaycy Lower Control Limit:
yayaya = yc 3yuaycy QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
By applying control limits, one can determine whether a process is in control.
A process is considered stable if plotted points remain within the control limits and follow a random pattern.
Points that fall outside these limits or form non-random patterns suggest the presence of special causes that require investigation and correction.
It is important to note that u-charts assume data follows a Poisson distribution and that defect rates are moderate.
Extremely low defect rates may require large sample sizes, while very high rates may limit the chartAos applicability.
Control charts like the u-chart are invaluable for monitoring process performance over time.
They typically feature three lines: a central line representing the average and upper and lower control limits, often set at three standard deviations.
Selecting an appropriate control chart is crucial, especially when dealing with attribute data such as defect counts.
After creating a control chart and interpreting its results, the next phase involves using the collected data to inform predictive models.
In this investigation, two modelsAiK-Nearest Neighbors (KNN) and Random Forest (RF)Aiwill analyze data from the u-chart to predict parameters of interest .
, air coil measurement.
These predictions help improve process understanding and guide informed decision-making.
Building Prediction Model After creating the control chart, the next step is to use the collected information to train a predictive model for the parameter of interestAisuch as air coil measurements.
In this study, two models will be employed: the K-Nearest Neighbors (KNN) algorithm and the Random Forest (RF) algorithm.
Both will use the data obtained from the u-chart to generate predictions, ultimately enhancing our understanding of the underlying process.
K-Nearest Neighbors K-Nearest Neighbors (KNN) is a supervised learning algorithm applicable to both classification and regression tasks.
In regression.
KNN identifies the k nearest data points relative to a test data point and then predicts the output as the average value of these k neighbors.
The prediction of KNN regression can be expressed as follows Rodriguez-Galiano, et al.
yc= yco
OO ( )
Here, the algorithm assumes that all input variables are on the same scale and that the target variable is continuous.
Choosing the optimal k is critical.
A larger k tends to yield smoother predictions, while a smaller k may provide more flexibility.
To determine the best k, itAos often useful to experiment with different values and evaluate performance using cross-validation or other error metrics.
Random Forest Random Forest regression is often employed for predicting continuous values rather than categorical outcomes.
As an ensemble method, it combines multiple decision trees to produce a final Each decision tree is trained on a randomly selected subset of the training data .
ith replacemen.
, and after all trees are trained, their predictions are averaged to yield the final result.
Equation 7 shows the calculation of predicted value of y (Rodriguez-Galiano, et al.
, 2.
The predicted output value for x is the average of the outputs predicted by all the trees:
ycIya.
= ycN yc .
Where:
x: input vector representing the set of features for which a prediction is to be made.
T: total number of individual decision trees that make up the Random Forest.
QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
: The prediction made by the tth decision tree in the Random Forest for the input x In this notation.
RF.
is the predicted output value for the input vector x.
Random Forest regression offers several advantages over other regression algorithms.
It is less prone to overfitting, can efficiently handle high-dimensional data, trains and predicts relatively quickly, and can model nonlinear relationships between input features and output values.
Its performance can be fine-tuned by adjusting hyperparameters such as the number of trees, the maximum tree depth, and the number of features considered at each split.
Optimal parameter values are often identified through crossvalidation (Alquthami, et al.
, 2.
In this study, a RF regression model was applied to the air coil scrap data to generate predictions and meet the research objectives.
Results and Discussion Results Model Selection Analysis Because the scope of the model is quite broad, a bar chart was used to pinpoint which models warrant closer attention.
A threshold, derived from the overall scrap data, serves as a benchmark for selecting the models under review.
This targeted approach simplifies the analysis, making it clearer and more aligned with the studyAos objectives.
Based on the production data, a threshold was established to identify which models require further analysis.
As shown in Figures 6 and 7, seven machines exceeded this thresholdAispecifically machines 1, 2, 4, 5, 10, 11, and 16.
The models involved are HA00-08464KLFTR.
HA00-17359ALFTR, and HM73-10C300LFTR.
Figure 6.
Threshold Productivity Model QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
Figure 7.
Productivity by Model U-Chart The utilization of the U-chart proves to be a viable means of monitoring the caliber of output from a CNC apparatus via pre-emptive maintenance.
The use of a chart aids in the detection of possible deficiencies before they develop into significant problems and contributes to the preservation of the machinery's output quality.
The utilization of CNC machine assists in gauging the caliber of the generated output and guaranteeing its adherence to the prescribed criteria of acceptability.
One limitation of this case study pertains to the temporal scope of the data, which is restricted to a one-year It can be argued that the control chart is more efficacious than plotting data by date or daily.
Through the graphical representation, it is possible to observe the progression of a certain procedure However, for greater specificity, a more detailed scrutiny can be carried out by analyzing the data daily.
The monitoring process will follow as in section 3.
8 to evaluate the u-chart.
But the evaluation focuses on the data that occur anomalies and in control at that month.
According to the data in Figure 8, only in January and May did the process remain consistently within the established control limits.
In contrast, significant volatility was observed during July and September, making process management challenging.
To pinpoint the problematic machine within the HA00-08464LFTR model, it is necessary to determine whether machine 1, machine 2, or machine 4 is responsible for the observed instability.
QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
Figure 8.
Control Chart for Model 1 According to the data in Figure 9, the process remained consistently within the established control limits during May, as indicated by the plot falling entirely within the designated control range.
However, certain periods, notably July and September, showed greater variability, posing challenges for effective process management.
Additionally, production ceased in November, coinciding with the conclusion of the Descriptive Analysis segment.
Figure 9.
Control Chart Model 2 As depicted in Figure 10, the process control remained within established limits only during October, as both periods displayed plots confined to the specified range.
In contrast, variability intensified in September, where a notable outlier emerged.
Despite this anomaly, the process remained stable, though it fell outside the expected parameters.
SeptemberAos scrap rate reached the highest recorded peak, posing additional challenges to maintaining effective process oversight.
QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
Figure 10.
Control Chart Model 3 Regression Model A forecasting analysis was conducted using data from 2022 on three models: Model 1 (HA0008464LFTR).
Model 2 (HA00-17359ALFTR), and Model 3 (HM73-10C300LFTR).
The machine learning pipeline remains consistent for all three, differing only in column identifiers.
Each modelAos data was split into training .
%) and testing .
%) sets to facilitate analysis.
Feature selection was then performed to identify the most relevant variables.
By removing unnecessary, redundant, or less informative features, feature selection helps streamline the model, improve accuracy, reduce overfitting, and shorten training times.
Next, a hyperparameter tuning approach was applied to optimize the models .
ee Figure .
Random search, a method for discovering the best set of hyperparameters, was employed.
Hyperparameters are preset values that influence a modelAos training behavior and overall performance.
By defining a parameter grid and conducting a grid search, the process finds the parameter combination that yields the best performance.
Random search tuning was applied to both RF and KNN models, previously selected during the model selection phase.
Figure 11.
Hyperparameter Tuning Code Figure 12.
Training Model Code QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
In this study, the chosen models are KNN and RF.
Among them.
RF is considered an ensemble model, while KNN is treated as a base model.
The evaluation criteria include minimizing the Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE), while maximizing the correlation coefficient (R) (Chicco, et al.
, 2.
Based on these criteria.
Table 1 indicates that the best-performing models are RF for Model 1 and Model 3, and KNN for Model 2.
Keyword Model 1 Model 2 Model 3 Table 1.
Evaluation for the best model
Model
ML Model
RMSE
HA00-08464LFTR
KNN
HA00-17359ALFTR
KNN
HM73-10C300LFTR
KNN
MAE
Although the modelAos performance was already established, further improvement can be achieved through stacking, an ensemble technique that layers multiple models to enhance predictive By blending predictions from different models, stacking can bolster overall performance.
shown in Table 2, most models benefit from stacking except for Model 1, where the RF model alone still outperforms the stacked approach.
To validate these findings, actual and predicted values were compared.
The actual data from January to March 2023 served as a reference, while the models were trained and tested exclusively on data from 2022 before forecasting and comparing the results against the 2023 actuals.
This approach ensures a realistic assessment of model performance in a true production environment.
Figure 13.
Stacking Method for Improvement of Model Keyword Model 1 Model 2 Model 3 Table 2.
Evaluation of Model by Adding Stacking Method
Model
ML Model
RMSE
MAE
HA00-08464LFTR
KNN
Stacking HA00-17359ALFTR RF
KNN
Stacking HM73-10C300LFTR RF
KNN
Stacking From Table 4, the stacking algorithm achieved the highest performance for Model 2 and Model 3, while Random Forest remained superior for Model 1.
Figures 14, 15, and 16 compare actual and predicted values produced by RF and the stacking approach.
Although RF reached higher peaks, the stacking predictions more closely matched the actual collected data.
Unlike testing on historical data.
QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
these graphs demonstrate how the models perform on real, current data, making the assessment more practical for real-world forecasting scenarios.
As a result, the stacking algorithm was selected to forecast the CNC scrap rate across the evaluated models.
Figure 14.
Actual VS Predicted Model 1 Figure 15.
Actual VS Predicted Model 2 Figure 16.
Actual vs Predicted Model 3 As shown in Figure 17, 18, and 19, the results indicate that the projected scrap rates for Models 1, 2, and 3 exhibit some irregularity.
This variability is partly due to excluding Sunday data, as the factory is typically shut down on that day.
Despite this introduced irregularity, the figure achieves the objective QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
of predicting product scrap rates.
The mean value for all models is below 0.
08, demonstrating both high accuracy and precision.
Such reliable forecasts can significantly support preventive maintenance efforts by helping prevent equipment anomalies or unexpected increases in scrap rates.
Figure 17.
Result Predicted Model 1 Figure 18.
Result Predicted Model 2 Figure 19.
Result Predicted Model 3 Percentage Different Error Between Actual and Predicted Data Table 3 indicates that Model 1Aos Mean Absolute Percentage Error (MAPE) was 75.
64%, suggesting its forecasts were off by an average of about three-quarters of the actual value.
Model 2 performed somewhat better with a MAPE of 42.
58%, but this still reflects a considerable deviation.
Model 3 yielded the most accurate forecasts, recording a MAPE of 39.
While this error rate is still high, it is notably lower than the other two models.
In summary.
Model 3 (HM73-10C300LFTR) was the most accurate of the three according to MAPE results.
Nonetheless, a 39.
35% error rate implies there is room for improvement.
Future work QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
could involve refining model parameters, experimenting with alternative algorithms, or incorporating additional variables to enhance prediction accuracy.
Table 3.
MAPE result using the Best Model
Keyword
Model
MAPE (%)
Model 1
HA00-08464LFTR
Model 2
HA00-17359ALFTR
Model 3
HM73-10C300LFTR
Table 4.
Data Actual and Predicted Date
1/1/2023
2/1/2023
3/1/2023
4/1/2023
5/1/2023
6/1/2023
7/1/2023
8/1/2023
9/1/2023
10/1/2023
11/1/2023
12/1/2023
13/1/2023
14/1/2023
15/1/2023
16/1/2023
17/1/2023
18/1/2023
19/1/2023
20/1/2023
21/1/2023
22/1/2023
23/1/2023
24/1/2023
25/1/2023
26/1/2023
27/1/2023
28/1/2023
29/1/2023
30/1/2023
31/1/2023
Model 1 Actual Model 2 Predicted Actual Model 3 Predicted Actual Predicted Dashboard for Monitoring the Data The dashboard, built in Power BI, transforms the Jupyter Notebook analysis into a visually engaging interface with filtering capabilities, offering fresh insights into the data.
As discussed in the method section, the u-chart is essential for monitoring process stability.
By integrating the u-chart into the dashboard, users can easily correlate process control information with other relevant datasets.
Figures 20, 21, and 22 illustrate the Power BI dashboard.
Card visualizations provide key metrics for all data, while filtering options allow users to focus on specific models or machines as needed.
The QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
line charts in Figures 21 and 22 utilize a basic control chart, including upper, center, and lower limits.
These limits apply to the entire dataset, not just a single model.
When filters are applied, the scrap rate is generally under control, though occasional anomalies still appear on certain dates.
Figure 20.
Dashboard HOME Figure 21.
Dashboard 2022 Figure 22.
Dashboard 2023 QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
Discussion Air coil output quality measurement can be obtained through the analysis of scrap data generated by the CNC machine.
At various instances, the occurrence of diverse air coil scrap is observed.
These varied scrap occurrences indicate a complex interplay of factors, including tool wear, machine calibration, material inconsistencies, or even operator skill levels.
Each factor potentially influences the final product quality .
Visualizing the scrap rates and production data via the dashboard provides a valuable lens through which anomalies and fluctuations become immediately apparent.
The u-chart, as a monitoring instrument, enables practitioners to detect variations beyond normal statistical limits (Ieren et al.
, 2020.
Tang et al.
, 2.
Such an analysis highlights the importance of establishing evidence-based benchmarks for identifying when the process deviates from its intended parameters.
However, while u-charts help reveal non-conformance, they do not inherently explain the root causes, thus requiring deeper diagnostic effortsAilike root cause analysis or additional process measurementsAito fully understand why anomalies persist.
When forecasting scrap rates, the decision to employ advanced machine learning modelsAiKNN and Random ForestAireflects an acknowledgment that simple statistical tools may be insufficient to capture the complexity of manufacturing data (Soori, et al.
, 2023.
Zhang & Jiang, 2.
Nevertheless, the modelsAo initial performance fell short of expectations, likely due to factors such as data quality, unmodeled process variability, or insufficient feature engineering.
In response, the study adopted a stacking approach, combining multiple predictive models to harness their individual strengths.
Although stacking notably improved performance for Models 2 and 3.
Model 1 did not exhibit the same This discrepancy may point to unique process conditions, variations in machine health, or parameter interactions not captured by the ensemble.
Choosing the stacking method over RF alone, even for Model 1, where RF showed higher peaks, was guided by stackingAos closer alignment with actual data.
This choice highlights a key challenge in predictive modeling: finding the right balance between pure predictive accuracy and stable, reliable Opting for the model that best matches real-world patterns, rather than one that is only statistically strong, demonstrates a practical approach to model selection.
Concluding that ensemble stacking is the preferred method for predicting scrap rates represents a move toward more advanced, data-driven preventive maintenance strategies.
It suggests that no single model type will always perform best and that combining various approaches can yield more robust result.
However, it is important to recognize that forecasting accuracy can be influenced by factors such as production schedule changes, maintenance timing, material quality, and evolving customer requirements.
While this study meets its primary goal of improving predictive capabilities, it also suggests the need for ongoing refinement.
Future research may incorporate additional data sourcesAilike sensor readings, operator logs, or environmental conditionsAito improve the modelsAo contextual Further enhancements could involve fine-tuning model parameters, revisiting the chosen features, or experimenting with more advanced algorithms like gradient boosting or neural In this way, these findings both confirm the value of ensemble methods in preventive maintenance and encourage further investigation into the complex factors affecting CNC machine performance and product quality.
Conclusion The study reveals the prospective application of u-charts and machine learning regression models to enhance preventive maintenance strategies for CNC machines.
By analyzing production and scrap data, we effectively predicted scrap rates, enabling more informed decision-making and resource The u-charts proved valuable for visualizing fluctuations in scrap rates over time, making it easier to identify patterns and anomalies.
Understanding these patterns led to a significant reduction in waste, which highlights the importance of data visualization in production management.
QOMARUNA Journal of Multidisciplinary Studies 2024.
Vol.
No.
01, pp.
Furthermore, the efficacy of regression models based on machine learning by using random forest (RF) and k-Nearest Neighbor .
NN) showed strong predictive capabilities.
Training these models with historical production data allowed us to anticipate scrap rates accurately, reduce downtime, and improve overall efficiency.
In addition, our study has revealed the complex dynamics relationship between u-charts and machine learning models.
U-charts helped identify areas needing attention, guiding model refinement by identifying key variables and unusual data points.
In turn, machine learning forecasts provided insights that could be monitored and tracked through u-charts, further improving process oversight.
This studyAos findings are constrained by several limitations.
The results depend on high-quality, comprehensive data and may not generalize to other machines or contexts.
Statistical assumptions about distributions and independence may not hold, and model selection was limited.
Practical factors like cost, scheduling, and training were not fully addressed.
Future research should explore more advanced machine learning methods, apply the approach in varied settings, incorporate real-time analytics, and consider broader factors such as economic constraints and operator skills.
This would enhance reliability, applicability, and overall understanding of predictive maintenance strategies.
To summarize, the integration of these methodologies into the preventive maintenance plan of CNC machines has demonstrated advantages in optimizing manufacturing activities, enhancing the standard of the final product, and minimizing material waste.
The present study highlights the prospects of utilizing data analysis and forecasting tools to transform the maintenance tactics implemented in the manufacturing sector.
Acknowledgment I would like to express my gratitude to my senior employees.
Encik Faizul.
Encik Zul.
Puan Hajar, and Puan Hayati who assisted and encouraged me all the way through the process of finishing this project
as well as all the years that I spent studying Declaration of Conflict of Interest
The authors declare no potential conflicts of interest related to the research, writing, and/or publication of this article References