INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

Data Visualization and Fault Detection Model for Sputtering Process

Monitoring

1,3

4

Ngoi Yuk Loong^1,2, Kok Swee Leong^1,2, Syafeeza Ahmad Radzi , Tong Song Pui

*

¹Faculty of Electronic and Computer Technology and Engineering, Universiti Teknikal Malaysia

Melaka, Hang Tuah Jaya, 76100 Durian Tunggal, Melaka, Malaysia

²Advanced Sensors and Embedded Controls System (ASECs), Centre for Telecommunication Research

and Innovation (CeTRI), Universiti Teknikal Malaysia Melaka, Hang Tuah Jaya, 76100 Durian Tunggal,

Melaka, Malaysia

³Machine Learning and Signal Processing (MLSP), Centre for Telecommunication Research and

Innovation (CeTRI), Universiti Teknikal Malaysia Melaka, Hang Tuah Jaya, 76100 Durian Tunggal,

Melaka, Malaysia

⁴Texas Instruments Electronics Malaysia, Free Trade Zone, Batu Berendam, 75350, Melaka, Malaysia

*

Corresponding Author

DOI: https://doi.org/10.47772/IJRISS.2025.91200053

Received: 09 December 2025; Accepted: 16 December 2025; Published: 31 December 2025

ABSTRACT

This paper introduces a real-time system for data visualization and fault detection specifically tailored for

industrial sputtering processes, focusing on monitoring parameters such as deposition rate, film thickness,

material types (Ti, Ag, Ni), and overall process status. The project's primary goals are to model multi-level

anomaly outliers to detect potential OCR errors and process deviations, implement a real-time embedded vision

system for automated data acquisition from the equipment's display (achieving high accuracy with YOLO and

PaddleOCR), and deploy a complete monitoring application featuring real-time visualization and post-process

analysis. Data acquisition is carried out using a high-resolution camera, where YOLO achieves 99.5% mAP@0.5

in supervised detection of visual indicators, and PaddleOCR attains 99.57% accuracy in extracting numerical

parameters. Preprocessing incorporates a median filter to suppress noise, while DBSCAN identifies sudden OCR

fluctuations and linear regression models parameter trends. The postprocessed data are stored in structured CSV

files. By integrating robust supervised and unsupervised learning techniques with data science methodologies,

the proposed solution ensures reliable operational monitoring, enables early anomaly detection, and supports

predictive maintenance strategies in industrial settings.

Keywords: Machine learning, Sputtering process Monitoring, Real-time Fault detection, Embedded Vision,

Optical Character Recognition (OCR), Anomaly Detection

INTRODUCTION

In the era of Industry 4.0, rapid and reliable sensor data acquisition is essential for predictive maintenance,

anomaly detection, and process optimization in modern manufacturing systems [1][2]. In semiconductor and

thin-film coating applications, sputtering processes require precise monitoring of critical parameters such as

electrical power, deposition rate, and film thickness. These processes often involve materials like titanium (Ti),

silver (Ag), and nickel (Ni), each exhibiting unique deposition characteristics. However, many legacy sputtering

machines lack built-in remote data access, making hardware upgrades difficult or impractical.

This paper presents a vision-based monitoring system that captures process parameters directly from sputtering

machine display panels using YOLO-based object detection [3][4] integrated with Optical Character

Page 606

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

Recognition (OCR). The system leverages the computational capability of the NVIDIA Jetson Orin Nano to

perform real-time image processing and efficiently extract key operational data. First, YOLO detects and

classifies visual indicators on the display, after which OCR extracts numerical parameters such as electrical

power, deposition rate, and film thickness. To enhance data reliability, median filtering is applied to suppress

noise, followed by the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm for

anomaly detection. This helps distinguish genuine process faults from abrupt OCR fluctuations. Linear

regression is then used to model deposition trends, enabling fault identification and trend analysis. All extracted

and time-stamped data are stored n structured CSV files, supporting both real time fault detection and data-

driven decision making.

The key contributions of this research are threefold: (1) the development of an embedded, non-invasive vision

system for real-time data acquisition from sputtering equipment; (2) the implementation of a multi-stage anomaly

detection framework that integrates OCR error screening with machine fault modeling; and (3) a comprehensive

performance evaluation of the detection models, OCR engines, and anomaly detection methods to establish

benchmarks for accuracy, efficiency, and robustness.

Problem Statement

A major challenge in modern manufacturing is the difficulty of obtaining real-time process data from legacy

equipment, where hardware modification is often impractical, costly, or risk disrupting machine performance.

This work addresses this issue in sputtering systems, which typically lack direct digital access to critical sensor

information. To overcome this limitation, this work proposes a non-invasive monitoring approach deployed on

an NVIDIA Jetson embedded platform that uses Optical Character Recognition (OCR) to extract process

parameters directly from the equipment's display panels. However, reliance on a vision-based method introduces

an inherent risk: inaccurate OCR readings may trigger false alarms or, conversely, mask actual machine faults.

Therefore, a central component of this research is the integration of robust data validation and error-handling

mechanisms to ensure the reliability and integrity of the captured data. This enables accurate real-time

visualization and supports proactive fault detection.

LITERATURE REVIEW

1. Overview of Sputtering Deposition Processes

Sputtering is a widely adopted Physical Vapor Deposition (PVD) technique used for fabricating thin films with

precise control over composition, thickness, and uniformity. The process is based on momentum transfer, in

which energetic ions, typically generated from an argon plasma, bombard a solid target material. As a result of

these collisions, atoms are ejected from the target surface and subsequently condense on a substrate, forming a

thin film, as illustrated in Figure 1. This mechanism, commonly referred to as cathode sputtering, supports the

deposition of metals, alloys, semiconductors, and dielectric materials with strong adhesion and uniform coverage

[5][6].

Figure 1: Physical sputtering processes [6]

Page 607

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

Several sputtering configurations have been developed to accommodate different materials and process

requirements. DC diode sputtering employed a constant voltage between the cathode and anode and is primarily

suitable for conductive targets. RF sputtering extends the technique to insulating materials by alternating the

electric field to prevent charge accumulation. Magnetron sputtering, shown in Figure 2, further enhances plasma

density through magnetic confinement, leading to higher deposition rates, improved target utilization, and better

film quality [7][8][6]. Due to these advantages, magnetron sputtering is widely used in semiconductor and

industrial coating applications.

Figure 2: Magnetron sputter deposition systems [6].

2. Critical Process Parameters for Sputtering Process Monitoring

Effective monitoring of sputtering processes relies on continuous observation of key operational parameters that

directly influence film quality and process stability. Modern sputtering systems typically display real-time values

for electrical power, deposition rate, accumulated film thickness, and overall process state. Together, these

parameters provide a comprehensive representation of system behavior during deposition [9].

Electrical power, measured in watts (W) or kilowatts (kW), determines the ion bombardment energy and directly

affects the sputtering yield. Stable power delivery is essential for consistent film growth, while power

fluctuations may indicate plasma instability, target degradation, or electrical faults.

The deposition rate represents the speed at which material accumulates on the substrate and is commonly

expressed in angstroms per second (Å/s) or nanometers per minute (nm/min). This parameter is influenced by

applied power, chamber pressure, target condition, and plasma density. Real-time monitoring of deposition rate

enables early detection of abnormal process behavior and facilitates process optimization.

Film thickness corresponds to the cumulative deposited layer and is obtained by integrating the deposition rate

over time. It is typically expressed in angstroms (Å/s) or nanometers (nm). Accurate thickness monitoring is

crucial to meeting design specifications and ensuring timely process termination [10]. Table 1 summarizes the

key process parameters commonly monitored in sputtering systems.

Table 1: Key Process Parameters of the Sputtering Machine

Parameter

Units

Typical Range

0-100 %

Power

Percentage (%)

Angstrom per second (Å/s)

Angstrom (kÅ)

Text/Number

Deposition Rate

Film Thickness

Process State

0.1-50 Å/s

0-3 kÅ

Various states

Page 608

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

3. Data Visualization for Industrial Process Monitoring

Data visualization plays a vital role in industrial process monitoring by transforming raw numerical data into

interpretable visual representations. In manufacturing environments, real-time visualization of process

parameters enables operators to quickly identify trends, deviations, and abnormal behavior that may otherwise

go unnoticed in numerical displays alone.

In sputtering systems, graphical visualization of power, deposition rate, and thickness profiles can reveal gradual

drifts, sudden spikes, or irregular fluctuations associated with process faults. Prior studies have demonstrated

that visual dashboards improve situational awareness, reduce operator response time, and support data-driven

decision-making in advanced manufacturing systems [11].

4. Optical Character Recognition for Industrial Data Acquisition

Optical Character Recognition (OCR) is a key enabling technology for extracting textual and numerical

information from visual sources, such as equipment displays and control panels. While OCR has been

extensively studied in applications like document digitization and Automatic License Plate Recognition (ALPR),

recent research has increasingly focused on its application in industrial monitoring and automation [12][13].

In industrial contexts, OCR enables non-invasive data acquisition from legacy equipment that lacks digital

communication interfaces. Kim et al. [16] demonstrated a low-cost OCR-based system using Tesseract to extract

real-time operational data from CNC machine displays. Similarly, Kaur et al. [17] proposed an Industrial Internet

of Things (IIoT) architecture incorporating OCR, achieving recognition accuracy above 97% under controlled

conditions.

However, OCR performance remains sensitive to environmental factors such as illumination variation, display

reflections, font diversity, and image noise [14]. Leaning-based OCR methods, including neural-network-based

classifiers, have been shown to outperform traditional template-matching approaches by adapting to diverse

character styles and imaging conditions [12][15]. These findings support the feasibility of OCR as a reliable data

acquisition method for real-time sputtering process monitoring.

5. Fault Detection and Anomaly Detection in Sputtering Processes

Fault detection in sputtering processes is essential for maintaining film quality, reducing material waste, and

preventing equipment damage. Traditional approaches often rely on rule-based thresholds or operator expertise,

which may fail to detect subtle or evolving faults. Consequently, data-driven anomaly detection techniques have

gained increasing attention in recent literature.

Clustering-based methods such as Density-Based Spatial Clustering of Applications with Noise (DBSCAN) are

well suited for identifying abnormal process behavior without requiring labeled fault data [18]. DBSCAN can

distinguish normal operating clusters from noise points representing potential faults. However, quantitative

performance evaluation is necessary to assess detection reliability. Common statistical metrics used in anomaly

detection include precision, recall, F1-score, and false alarm rate, which measure the accuracy and robustness of

fault identification.

Regression-based models, such as linear regression, are frequently employed as baseline predictors of expected

process behavior. Deviations between predicted and observed values can be interpreted as anomalies. To ensure

meaningful interpretation, confidence intervals, residual analysis, and statistical thresholds are typically applied

to quantify deviation significance. Prior studies emphasize that combining visualization with statistically

validated anomaly detection enhances interpretability and operator trust in intelligent monitoring systems.

METHODOLOGI

1. System Overview

The proposed monitoring system follows a structured workflow consisting of data acquisition, data

Page 609

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

preprocessing, and data modeling, as illustrated in Figure 3. The system is designed to operate in a non-invasive

manner by leveraging computer vision and data-driven analytics to monitor critical sputtering process parameters

in real time. The primary parameters of interest include electrical power, deposition rate, and film thickness.

Figure 3: Flow Chart

2. Data Acquisition and Visual Parameter Localization

Real-time images of the sputtering machine’s display panel are captured using a high-resolution industrial

camera mounted at a fixed position to ensure consistent viewing geometry. The camera continuously records

images under normal operating conditions, capturing numerical indicators corresponding to key process

parameters.

To localize and identify these indicators, the YOLO (You Only Look Once) object detection framework is

employed as shown in Figure 4. YOLO is selected due to its real-time inference capability and high detection

accuracy. The model detects predefined regions of interest (ROIs) corresponding to electrical power, deposition

rate, and film thickness displays.

Page 610

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

Figure 4: (a) Screenshots of sputtering machine display panel with YOLO detection outputs, (b) illustration of

bounding box localization and class labels for each monitored parameter, (c) data presented in real-time graph.

3. Dataset Preparation and Model Training

A custom dataset was constructed specifically for this study. The dataset was divided using an 80:10:10 ratio,

consisting of 5046 training images, 644 validation images, and 656 testing images, as shown in Figure 5.

Transfer learning was applied using pre-trained YOLO weights to accelerate convergence and improve

generalization. Fine-tuning was conducted on the custom dataset to adapt the model to the specific visual

characteristics of the sputtering machine display. The training configuration is summarized in Table 1.

Figure 5: Data Split (Train, Validation, Test)

Model performance was evaluated using standard object detection metrics, including precision, recall, and mean

Average Precision (mAP@0.5), to ensure robust and accurate detection prior to deployment.

Table 1: YOLO Training Configuration Parameters

Training Parameters

Task specification

Model

Parameters

detect

yolo11s.pt

Dataset location

{dataset.location}/data.yaml

Page 611

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

Number of epochs

100

stImage size (imgsz)

Batch size

640, 480

16

Optimizer

AdamW

002

Learning rate

Early stopping

Weight decay

10

0.005

4. OCR-Based Data Extraction and Preprocessing

Following object detection, the localized ROIs are passed to the Optical Character Recognition (OCR) module

for numerical value extraction. OCR processing includes a preprocessing pipeline designed to improve

recognition accuracy under varying illumination and display conditions. The preprocessing steps include

conversion to grayscale, adaptive thresholding to enhance text contrast, noise reduction using median filtering,

and morphological operations to refine character boundaries.

Tesseract OCR is employed for numerical recognition, with customized character whitelists and confidence

thresholds to reduce misclassification. OCR outputs with confidence scores below a predefined threshold are

discarded or flagged for reprocessing.

5. Anomaly Detection Using DBSCAN

To detect abrupt abnormalities in the sputtering process, Density-Based Spatial Clustering of Applications with

Noise (DBSCAN) is applied to the OCR-extracted numerical data. DBSCAN is well suited for this application

due to its ability to identify outliers without requiring labeled fault data.

5.1. Parameter Selection for DBSCAN

The DBSCAN parameter ε (epsilon) and minPts were selected based on empirical analysis and domain

knowledge. The ε value was determined using a k-distance plot, where the knee point indicates a suitable

neighborhood radius. The minPts parameter was set to 4, following common practice for low-dimensional

industrial datasets.

These parameters enable DBSCAN to form dense clusters representing normal operating conditions while

labeling isolated points as anomalies.

5.2. Statistical Evaluation of Anomaly Detection

Anomaly detection performance was quantitatively evaluated using precision, recall, and F1-score. Ground truth

labels were established through expert inspection of process logs and corresponding visual trends. Precision

measures the proportion of correctly identified anomalies, while recall evaluates the system’s ability to detect

actual faults. The F1-score provides a balanced assessment of detection accuracy.

6. Trend Analysis Using Linear Regression

In addition to detecting abrupt faults, linear regression models are employed to monitor long-term trends in

process parameters such as deposition rate and film thickness. Regression models are trained on historical

normal-operation data to establish baseline behavior.

Deviation from predicted values is assessed using residual analysis. Confidence intervals are computed to

Page 612

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

determine statistically significant deviations from expected trends. When residuals exceed predefined thresholds,

gradual process drifts or performance degradation are flagged.

Regression performance is evaluated using metrics including Mean Absolute Error (MAE), Root Mean Square

Error (RMSE), and coefficient of determination (R²).

7. System Deployment and Real-Time Monitoring

During deployment, the system operates continuously by capturing display images, localizing parameter regions

using YOLO, extracting numerical values via OCR, and performing anomaly detection using DBSCAN and

regression analysis. The integrated system provides real-time alerts for both sudden anomalies and gradual drifts

without requiring any modifications to the exiting sputtering hardware.

RESULTS AND DISCUSSION

1. YOLO Model Performance for Visual Indicator Detection

Table 2 summarizes the detection performance of multiple YOLO variants evaluated for identifying visual

indicators on the sputtering machine’s display panel. All evaluated models achieved precision (1.0) and recall

(1.0), indicating consistent and reliable localization of the predefined regions of interest.

Among the tested models, YOLOv8s achieved a mean Average Precision of 99.5% (mAP@0.5) and

demonstrated a favorable balance between detection accuracy and computational efficiency. Although

YOLOv5s and YOLOv12s exhibited slightly higher mAP@0.5:0.95 values, YOLOv8s maintained comparable

accuracy with a moderate model size of 21.4 MB and reduced architectural complexity (164 layers). These

characteristics make YOLOv8s well suited for real-time deployment on embedded or resource-constrained

industrial systems.

Based on detection accuracy, inference efficiency, and deployment feasibility, YOLOv8s was selected as the

optimal object detection backbone for the proposed monitoring system.

Table 2: YOLO Model Performance Comparison

Model

Precision

Recall

mAP@0.5

0.995

mAP@0.5:0.95

0.915

Parameters (M)

Layers

193

YOLOv5s

YOLOv8s

YOLOv11s

YOLOv12s

1

9.11

11.13

9.41

9.08

0.995

0.907

168

0.995

0.907

238

0.995

0.915

376

2. OCR Performance Evaluation

To evaluate OCR accuracy, 1,000 uniformly sampled frames were extracted from the source video and processed

using PaddleOCR, EasyOCR, and PyTesseract. Table 3 presents the quantitative comparison.

PaddleOCR achieved the highest recognition accuracy at 99.9%, significantly outperforming EasyOCR and

PyTesseract, which exhibited substantial misclassification under industrial display conditions. The superior

performance of PaddleOCR is attributed to its robust deep-learning-based text recognition architecture, which

effectively handles font variation, illumination changes, and display noise.

These results confirm that PaddleOCR is the most reliable OCR engine for extracting numerical parameters from

sputtering machine display panels.

Page 613

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

Table 3: OCR Performance Comparison for 1,000 Video Frames

OCR Tools

Correctly read values Incorrectly read Accuracy (%)

Error (%)

by OCR

values by OCR

PaddleOCR

EasyOCR

999

1

99.9

48.3

55.3

0.1

483

517

447

51.7

44.7

PyTesseract

553

3. Statistical Validation of OCR Anomaly Detection Using DBSCAN

DBSCAN was applied to the OCR-extracted numerical values to identify abrupt anomalies indicative of OCR

errors or process disturbances. Ground truth anomaly labels were established through manual inspection of time-

series plots and corresponding video frames.

Using these labels, anomaly-detection performance was quantitatively evaluated using precision, recall, and F1-

score, defined as:

 Precision: proportion of detected anomalies that are true anomalies

 Recall: proportion of true anomalies correctly detected

 F1-score: harmonic mean of precision and recall

DBSCAN achieved high anomaly-detection performance, with precision exceeding 95% and recall above 90%

across the evaluated dataset. This indicates that DBSCAN effectively identifies true anomalies while minimizing

false alarms

Figure 6 illustrates a representative DBSCAN result, where a sudden and unexpected drop in OCR values around

frame 1,500 was correctly identified as an outlier (highlighted in red). After the anomaly, the signal returned to

the expected operational trend, confirming DBSCAN’s robustness in detecting abrupt non-linear deviations.

Figure 6: DBSCAN Anomaly Detection on Value

4. Linear Regression-Based Machine Fault Detection

Linear regression models were constructed for three sputtering sources – Titanium (Ti), Silver (Ag), and Nickel

(Ni) – to characterize deposition behavior under stable operating conditions. Figure 8 compares the fitted

regression models.

Page 614

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

The estimated deposition rates confirm the material-dependent nature of sputtering process. Silver (Ag) exhibited

the highest deposition rate (0.0351 kÅ/s), followed by Titanium (Ti) at 0.0245 kÅ/s and Nickel (Ni) at 0.0060

kÅ/s.

Figure 8: Comparison of Linear Regression Models (TI, AG, NI sources)

The fitted regression equations are expressed as:

TI source:

(1)

(2)

(3)

Thickness (kÅ) = 0.0634 + 0.024526 × Time (s)

AG source:

Thickness (kÅ) = 0.0972 + 0.035119 × Time (s)

NI source:

Thickness (kÅ) = 0.0126 + 0.006009 × Time (s)

The general deposition model is defined as:

Thickness = 휷_ퟎ+ Rt

where, β₀represents the intercept, R denoted the deposition rate, and t is time.

Empirical analysis indicates that β₀is negligible compared to linear growth term, allowing the model to be

simplified as:

Thickness  Rt

Regression model performance was statistically evaluated using the coefficient of determination (R²), Mean

Absolute Error (MAE), and Root Mean Square Error (RMSE). High R²value (greater than 0.98 for all three

sources) indicates strong linearity and high prediction reliability. Confidence intervals computed for the

regression slopes proved statistical bounds for acceptable deposition-rate variation. Deviations beyond these

confidence limits are treated as indicators of potential machine faults or process drift.

Page 615

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

INTEGRATED DISCUSSION

The experimental results validate the effectiveness of the proposed embedded vision-based sputtering process

monitoring system across all stages of the pipeline.

YOLO-based object detection demonstrated consistently high accuracy, ensuring reliable localization of process

parameters in real-time. PaddleOCR provide near-perfect numerical extraction accuracy, significantly

outperforming alternative OCR engines in industrial display conditions. DBSCAN achieved high precision and

recall in detecting OCR anomalies, confirming its suitability for unsupervised fault detection. Linear regression

models established statistically validated baseline deposition behavior, enabling detection of gradual machine

faults through confidence-based deviation analysis.

Overall, the integrated system exhibits strong robustness, statistical reliability, and real-time capability, making it

suitable for deployment in industrial sputtering environments. By combining vision-based data acquisition with

quantitatively validated anomaly-detection techniques, the system enhances process stability, fault detection, and

predictive maintenance effectiveness.

CONCLUSION

This paper presented a non-invasive, vision-based monitoring framework for real-time sputtering process supervision,

integrating object detection, optical character recognition, data visualization, and statistically validated fault detection models.

The proposed system enables continuous extraction and analysis of critical sputtering parameters including electrical power,

depositionrate, and filmthickness, without requiring additionalsensorsor modificationsto existing equipment.

Experimentalresultsdemonstratethat YOLO-basedobject detection provides highlyreliable localizationofprocess indicators,

while PaddleOCR achieves near-perfect numerical extraction accuracy under industrial display conditions. The integration of

DBSCANenables effectivedetectionofabruptanomalies inOCR-deriveddata, supportedbyquantitativeperformancemetrics

such as precision, recall, and F1-score. In parallel, linear regression models establish statistically robust baseline deposition

behaviorfordifferentsputteringmaterials,withconfidence-baseddeviationanalysisenablingearlydetectionofgradualmachine

faults andprocess drift.

The proposed framework offers significant advantages for industrial manufacturing environments, including low deployment

cost, compatibility with legacy equipment, and real-time fault awareness. By combining data visualization with statistically

validatedanomalydetection,thesystemenhancesprocessstability,reducestheriskofundetectedfaults, andsupportspredictive

maintenance strategies in sputtering operations.

Despite itseffectiveness,thecurrent studyfocusesonalimitedsetofmaterials andcontrolledoperatingconditions. Futurework

will extend the framework to additional sputtering sources, multi-target systems, and more complex deposition recipes. The

integrationofadvancedpredictive models andadaptivethresholding mechanisms will further improve fault diagnosis accuracy

and scalabilityin high-volume manufacturing environments.

ACKNOWLEDGEMENT

The authors would like to acknowledge the support of this work by University Teknikal Malaysia Melaka

(UTeM) and Texas Instruments Electronics Malaysia, on short term student’s research grant

“INUDSTRI/TEXASINSTRUMENTS/FTKEK/2025/IO101.

REFERENCES

1. J. Yan, Y. Meng, L. Lu and L. Li, "Industrial Big Data in an Industry 4.0 Environment: Challenges,

Schemes, and Applications for Predictive Maintenance," in IEEE Access, vol. 5, pp. 23484-23491, 2017,

doi: 10.1109/ACCESS.2017.2765544.

2. F. M. Abdelillah, H. Nora, O. Samir and S. M. Benslimane, "Predictive Maintenance Approaches in

Industry 4.0: A Systematic Literature Review," 2023 IEEE International Conference on Enabling

Technologies: Infrastructure for Collaborative Enterprises (WETICE), Paris, France, 2023, pp. 1-6, doi:

Page 616

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025

10.1109/WETICE57085.2023.10477842

3. M. Ponika, K. Jahnavi, P. S. V. S. Sridhar and K. Veena, "Developing a YOLO based Object Detection

Application using OpenCV," 2023 7th International Conference on Computing Methodologies and

Communication (ICCMC), Erode, India, 2023, pp. 662-668, doi: 10.1109/ICCMC56507.2023.10084075

4. N. P, M. S. K, A. M. Gurusigaamani, B. S. Vasanth, V. S. S. Saranya and P. G. K. Reddy, "Traffic

Detection and Enhancing Traffic Safety: YOLO V8 Framework and OCR for Violation Detection Using

Deep Learning Techniques," 2024 International Conference on Science Technology Engineering and

Management (ICSTEM), Coimbatore, India, 2024, pp. 1-6, doi: 10.1109/ICSTEM61137.2024.10560904

5. F. M. Mwema, T. Jen, and L. Zhu, “Thin Film Coatings”. 2022. doi: 10.1201/9781003202615.

6. “Handbook of Sputtering Technology,” 2012, doi: 10.1016/c2010-0-67037-4.

7. Jilani, M. Sh. Abdel-wahab, and A. H. Hammad, “Advance Deposition Techniques for Thin Film and

Coating,” in InTech eBooks, 2017. doi: 10.5772/65702.

8. Shahzad, Plasma Science and Technology. IntechOpen, 2021. doi: 10.5772/intechopen.95256.

9. R. P. D. Nardo, P. Z. Takacs, C. F. Majkrzak, and P. M. Stefan, “Sputter deposition system for controlled

fabrication of multilayers,” OSTI OAI (U.S. Department of Energy Office of Scientific and Technical

Information), Jun. 1985, Accessed: Jan. 2025. [Online]. Available: https://www.osti.gov/biblio/5205090 .

10. Buzea and K. Robbie, “State of the art in thin film thickness and deposition rate monitoring sensors,”

Reports on Progress in Physics, vol. 68, no. 2, p. 385, Jan. 2005, doi: 10.1088/0034-4885/68/2/r04.

11. Al-Sulaiti, M. Mansour, H. Al-Yafei, S. Aseel, M. Kucukvar, N. C. Onat, ”Using Data Analytics and

Visualization Dashboard for Engineering, Procurement, and Construction Project’s Performance

Assessment”, IEEE 8^thInt. Conf. on Industrial Engineering and Applications (ICIEA), April 2021,

Chengdu, China.

12. -N. E. Anagnostopoulos, I. E. Anagnostopoulos, I. D. Psoroulas, V. Loumos and E. Kayafas, "License

Plate Recognition From Still Images and Video Sequences: A Survey," in IEEE Transactions on

Intelligent

Transportation

Systems,

vol.

9,

no.

3,

pp.

377-391,

Sept.

2008,

doi:

10.1109/TITS.2008.922938.

13. S. Mori, C. Y. Suen and K. Yamamoto, "Historical review of OCR research and development," in

Proceedings of the IEEE, vol. 80, no. 7, pp. 1029-1058, July 1992, doi: 10.1109/5.156468.

14. S. Du, M. Ibrahim, M. Shehata and W. Badawy, "Automatic License Plate Recognition (ALPR): A State-

of-the-Art Review," in IEEE Transactions on Circuits and Systems for Video Technology, vol. 23, no.

2, pp. 311-325, Feb. 2013, doi: 10.1109/TCSVT.2012.2203741.

15. L. Neumann and J. Matas, "Real-time scene text localization and recognition," 2012 IEEE Conference

on Computer Vision and Pattern Recognition, Providence, RI, USA, 2012, pp. 3538-3545, doi:

10.1109/CVPR.2012.6248097

16. H. Kim, W.-K. Jung, I.-G. Choi, and S.-H. Ahn, "A low-cost vision-based monitoring of computer

numerical control (CNC) machine tools for small and medium-sized enterprises (SMEs)," Sensors, vol.

19, no. 20, p. 4506, Oct. 2019.

17. R. Kaur, J. Acharya, and S. Gaur, "Non-invasive data extraction from machine display units using video

analytics," World Academy of Science, Engineering and Technology, International Journal of Computer

and Information Engineering, vol. 13, no. 7, pp. 419–424, 2019.

18. Deng, “DBSCAN Clustering Algorithm Based on Density”, 7^thInt. Forum on Electrical Engineering and

Automation (IFEEA), September 2020, Hefei, China.

Page 617

www.rsisinternational.org