processes

Review
Capability Indices for Digitized Industries: A Review and
Outlook of Machine Learning Applications for Predictive
Process Control
Jan Mayer * and Roland Jochem

Faculty of Mechanical Engineering and Transport Systems, Technische Universität Berlin, Pascsalstr. 8-9,
10587 Berlin, Germany; [email protected]
* Correspondence: [email protected]

Abstract: Leveraging machine learning applications for predictive process control signifies a decisive
advancement in manufacturing quality management, transitioning from traditional descriptive to
predictive capability indices. This review highlights the growing importance of predictive process
control, essential for quality assurance and the dynamic adaptability of production lines, which is
paramount in satisfying stringent quality standards and evolving consumer demands. The investiga-
tion into the integration of comprehensive sensor networks and sophisticated algorithmic analytics
enriches continuous improvement strategies, markedly enhancing the accuracy and efficiency of
production quality monitoring and control mechanisms. By moving beyond the limits of statistical
process control to predictive methods enabled by machine learning algorithms, the study presents
a transformative leap in manufacturing processes. The presented findings illustrate the critical role of
predictive algorithms in navigating the complexities of process variability, thereby ensuring consistent
adherence to established quality specifications. This approach not only facilitates immediate and
accurate product quality categorization, increasing overall operational efficiency, but also equips
manufacturers to swiftly respond to the variable nature of manufacturing requirements. Furthermore,
this research delves into the multifaceted impacts of predictive process control on the manufacturing
ecosystem. The ability to predict process quality decrease before it occurs, the optimization of resource
allocation, and the anticipation of production bottlenecks before they impact output are among the
notable benefits of this technological evolution. These developments to predictive process control
Citation: Mayer, J.; Jochem, R.
is instrumental in propelling the manufacturing industry toward a more agile, sustainable, and
Capability Indices for Digitized
Industries: A Review and Outlook of
customer-centric future. This shift not only complements the industry’s drive toward comprehensive
Machine Learning Applications for digitization but also promises significant strides in achieving superior process improvements and
Predictive Process Control. Processes maintaining a competitive edge on the global market.
2024, 12, 1730. https://doi.org/
10.3390/pr12081730 Keywords: predictive process control; statistical process control; machine learning; process
management; quality management
Academic Editor: Jie Zhang

Received: 27 June 2024

Revised: 2 August 2024
Accepted: 14 August 2024 1. Introduction
Published: 16 August 2024
The digital era has profoundly reshaped the landscape of manufacturing industries.
At the core of these changes is the imperative for businesses to adapt to escalating consumer
demands and to the principles of continuous improvement management. The challenge
Copyright: © 2024 by the authors.
that arises for manufacturers is the establishment of production processes that are both
Licensee MDPI, Basel, Switzerland. robust and efficient [1]. When the stringent requirements of these processes are not met,
This article is an open access article there often is a necessity for physical intervention which can compromise the stability
distributed under the terms and of these processes. Central to the decision making in process adjustments is the Process
conditions of the Creative Commons Capability Indicator (PCI). The PCI functions as an analytical benchmark, facilitating the
Attribution (CC BY) license (https:// evaluation of a process’s ability to produce output within predefined quality limits. It
creativecommons.org/licenses/by/ is a crucial determinant of whether a production process conforms to specific numerical
4.0/). quality thresholds, aiding in the assertion of quality adherence [2]. The evaluation of

Processes 2024, 12, 1730. https://doi.org/10.3390/pr12081730 https://www.mdpi.com/journal/processes

Processes 2024, 12, 1730 2 of 20

process capability necessitates a sequential documentation of the inherent quality criteria

of products during their manufacturing stage. This implies that continuous measurements
must be integrated within the production process, capturing data in a temporal sequence.
Such data are then analyzed using Statistical Process Control (SPC) methodologies to
compute the PCI. By examining the statistical positional parameters of a sample, a metric is
developed, from which inferences regarding the defect rate and overall process quality can
be extrapolated [3].
The rise of advanced data processing methods, particularly in the domain of Machine
Learning (ML), has had a substantial impact on this landscape. ML, a subset of artificial
intelligence, refers to the development of algorithms that enable computers to learn pat-
terns and make decisions based on data. Unlike traditional programming, where explicit
instructions are coded, ML algorithms learn from data inputs and can improve their
performance over time. This ability to learn and adapt makes ML particularly effective in
enhancing predictive decision making by leveraging large datasets. As the volume and
quality of data grow, so does the accuracy of statistical prediction algorithms, thereby
enabling a more refined approach to quality control and process optimization [4]. Further-
more, technological advancements are not limited to data acquisition; they also encompass
data transmission, enabling real-time data processing and facilitating immediate data
analysis for advanced decision making [5]. The implementation of such technologies
plays a pivotal role in ensuring ongoing monitoring of process quality and adherence to
customer specifications, thereby driving sustainable improvements in process reliability
and efficiency [6]. This method will hereafter be referred to as Predictive Process Control
(PPC), which involves the prediction of critical quality attributes for process control.
The structure of the paper is as follows: Section 2 presents a detailed review of existing
literature on PPC, emphasizing the integration of ML applications for enhancing manu-
facturing quality management. In Section 3, the methodology adopted for the systematic
literature analysis (SLA) is discussed, including the selection criteria, data sources, and
analytical approaches employed. Section 4 delves into the findings of the review, catego-
rizing the applications of ML in PPC and highlighting key trends and gaps in the current
research landscape. In Section 5, we engage in a critical discussion of the results, exploring
the implications for industry practice and future research directions. Section 6 concludes
the paper by summarizing the key insights and offering recommendations for leveraging
ML in PPC to achieve greater operational efficiency and product quality in manufacturing.

2. Background
The implementation of process quality key figures is being contradicted by these
advancements. Thus, the deployment of SPC, especially in the early stages of machine op-
eration subsequent to the conclusion of the design phase, to configure machine parameters,
represents a substantial instrument. This facilitates the continuous monitoring of quality
requirements after the start of serial production (SOP). After this indicated time step, only
statistically significant statements regarding the past can be made due to the descriptive
characteristics and the utilization of randomized samples [7].
Figure 1 illustrates the chronological arrangement of the SPC and the impact associated
with a decreasing number of acceptable parts, which directly influences the PCI and process
quality, respectively. During the start-up phase, the process quality gradually improves to
reach the desired tolerance level. At this juncture, series production begins, along with the
collection of samples for the SPC. During the second iteration of the SPC, the backlog of
acceptable parts falls within a range that causes the PCI to drop below the desired tolerance
level. This realization prompts physical intervention in the process, resulting in an increase
in the number of acceptable parts and, consequently, the process quality. However, it
is evident that due to the retrospective observation period, the SPC leads to a reactive
decision-making process, thus resulting in the production of avoidable missing parts.
This avoidability is primarily attributed to the utilization of industrial ML-applications,
which facilitate the advancement of decision support through the utilization of sensor
Processes 2024, 12, 1730 3 of 20

data. However, it enables targeted statements to be made regarding future product and
process quality. For instance, predictive algorithms are employed to make forecasts about
production processes or to classify products into different quality groups. As a result,
sustainable support processes are established to ensure compliance with the required
quality criteria [8,9].

Process quality Process quality

Produced within tolerance level not in
quantity tolerance level

Physical
SOP process
intervention

Quality confirm
Process
quality
products

Tolerance
level

Conception
Observation Observation
Phase period period

Start SPC SPC Time

Ramp-up Serial production

Figure 1. Chronological arrangement of the SPC in serial production.

Since the primary purpose of SPC is to provide statistically significant evidence of the
entire production process based on an inherent quality criterion, specific characteristics
related to the process are defined to calculate the quality of the process. The quality criterion
is associated with individual tolerance limits, namely the upper tolerance limit (UTL) and
the lower tolerance limit (LTL), which determine whether the manufactured workpiece
should be classified as rework or scrap if it exceeds or falls short of these limits [10]. To avoid
the cost-intensive measure of 100% quality control in series production, descriptive indices
such as process potential (C p ) and process performance (C pk ) are computed based on
random samples drawn to assess the quality characteristic of the products produced. This
allows for statistically significant conclusions to be drawn regarding compliance with
the tolerance limits for the entire population. Moreover, by using different sample sizes,
adjusted assessments of the influences of factors such as human, material, method, and
environment on the process can be made [11]. When a sample size of 50 is used for the
calculation, the defined metrics are machine capability (Cmk ) and machine potential (Cm ),
which assumes a less powerful statement about the capability. Herein, only 1 M (Machine) is
under control, whereas a higher sample size assumes more influence (5 M) is under control
(in addition to Machine: Man, Method, Material, and Milieu). In each case, the potential is
interpreted by comparing the tolerance limits to the spread. The capability indices further
relate the scatter width to statistical position parameters, enabling the determination of
both the process variation and the quantity of defective parts produced [2].
According to Dietrich and Conrad [12], the distinctive characteristic of the determined
indicator resides in both the temporal sampling of the sample and its size. Figure 2
presents a temporal relationship between the indicators and the level of confidence in
their enablement.
Processes 2024, 12, 1730 4 of 20

Machine capability Process capability

short-term provisionally long-term

Man
Method
Material
Machine
Machine Milieu

50 5 5 5 5 5 5 5 5
n ≥ 50 n ≥ 90 n ≥ 125

Requirement for Cm ≥ 2.0 Pp ≥ 1.67 C p ≥ 1.33

qualification Cmk ≥ 1.67 Ppk ≥ 1.67 C pk ≥ 1.33

Figure 2. Classification and characterization of capability indices, following Dietrich and Conrad [12].

For the determination of the capability indices, only single measured quality attributes
are utilized. Similarly, ML algorithms are designed with a single dependent variable,
which enables the transformation of previous descriptive capability indices into predictive
capability indices. Supervised algorithms, which leverage historical data to adapt to the
past behavior of dependent variables, can optimize learning behavior by expanding the
descriptive variables with sensor values. However, autoregressive models also allow for
univariate predictions. Unsupervised algorithms, predominantly used for pattern recogni-
tion, enable classifications based on defined clusters. Nonetheless, these classifications do
not establish a concrete affiliation and may indicate other dependencies within a cluster [13].
This interpretive limitation restricts the application of PPC. Given the diverse outcomes
of ML algorithms and the multilevel execution of SPC, supervised prediction algorithms
serve as the foundation for addressing the challenge of PPC.
In the context of digitized industries, various forecasting methods have been devel-
oped and applied to enhance process control and quality management. Table 1 provides an
overview of some of these existing forecasting methods, along with their descriptions and
relevant citations:

Table 1. Existing forecasting methods which can be used for PPC.

Forecasting Method Description

Utilizes historical data to predict future outcomes based on identified
Time Series Analysis patterns and trends in the data over time. Commonly used methods
include ARIMA and Exponential Smoothing [4].
Employs statistical techniques to model the relationship between
Regression Analysis a dependent variable and one or more independent variables,
enabling predictions based on this relationship [14].
Includes various algorithms such as neural networks, decision trees,
Machine Learning and support vector machines that learn from data to make
Models predictions, often outperforming traditional statistical methods in
complex datasets [15].
Combines multiple machine learning models to improve prediction
Ensemble Learning accuracy, including methods such as Random Forests and Gradient
Boosting Machines [8].
Involves creating models that simulate different scenarios to predict
Simulation-based
the impact of various factors on process outcomes, often used in
Forecasting
conjunction with ML for more accurate predictions [5].
Utilizes rule-based systems that mimic the decision-making abilities
Expert Systems of human experts to predict outcomes, often integrated with ML for
enhanced decision support [6].
Processes 2024, 12, 1730 5 of 20

These forecasting methods represent a blend of traditional statistical techniques and

modern machine learning approaches, each offering distinct advantages depending on the
specific requirements of the manufacturing environment.

2.1. Interpretation of ML Applications in the Context of PPC

As a consequence of the predominant focus of applied ML technologies on optimizing
decision processes and their outcomes, the interpretation and causation of predictive
elements have been increasingly overlooked [16]. This characterization has led to the
inclination to develop intricate algorithms known as black-box models, which only permit
limited inferences about algorithmic decision making. Only in recent years have endeavors
to develop interpretable models become discernible. These efforts are encompassed by the
comprehensive term “explainable Artificial Intelligence” (xAI) [17]. Given the importance
of interpreting the results, xAI methods contribute to an increase in information within
the context of process quality. In order to situate industrial ML applications in a suitable
environment, companies must consider not only information gains and knowledge transfer
but also various concepts for the industrial application of decision-supporting ML in
quality management for operational integration [18]. The primary focus lies in functionality,
although the economic benefit can only be partially measured and the integration costs
are a critical determinant of the technology’s success. With the shift towards digitized
corporate environments, production companies are confronted with the decision of which
decision-making processes should be supported by ML and what requirements should be
met by the application [19].
In order to be able to place industrial ML applications in an appropriate environment,
not only information gains and knowledge transfer play a decisive role for companies.
Therefore, a variety of concepts for the industrial application of decision-supporting ML
in quality management for operational integration is available [18]. The main focus is
on functionality, whereby the economic benefit can only be measured to a limited extent
and the integration costs are a decisive factor in the success of the technology. Due to
the industrial adaptation to a digitized corporate environment, production companies are
faced with the decision as to which decision-making processes are to be supported with
the help of ML and which requirements are to be met by the application [19].

2.2. Scope and Aim of the Study

The ultimate goal of an ML application is to continuously supervise the quality of the
production process by incorporating predictive elements. This empowers process managers
to make well-informed decisions prior to the occurrence of production errors, leading to
numerous advantages for the company. The decrease in defective parts not only reduces
production costs by minimizing the need for rework, but also optimizes the utilization
of resources that can be reassigned. Nonetheless, the deployment of ML applications in
the production setting involves multiple funtional units, including both employees and
machines. As depicted in Figure 3, the thematic relationships are visually presented.
In the area of manufacturing, the utilization of interpretable methods is of paramount
importance in order to fulfill the requisites for interactive prediction systems. To provide
recommendations to process owners, it is imperative to compare the interpretive capabilities
of ML methods for flexible designed models. Thus, the primary research question (RQ1)
pertains to the utilization of ML methods in the field of interactive prediction systems
in manufacturing and their interpretive capabilities for numerical, binary, and multiclass
predictions in contrast to descriptive models.
Moreover, the accuracy of predictive algorithms plays a pivotal role in ensuring the
high quality of models for numerical, binary, and multiclass prediction of the process
quality, as it directly affects the intervention decision. This raises the question as to whether
contemporary prediction algorithms, such as reinforcement learning and neural networks
such as Long Short-Term Memory (LSTM), offer advantages over other methods in this re-
gard. Hence, the second research question (RQ2) focuses on whether predictive algorithms
Processes 2024, 12, 1730 6 of 20

can exhibit high model quality and accuracy in numerical, binary, and multiclass prediction
of the C pk value, and whether modern prediction algorithms such as reinforcement learning
and LSTM provide advantages over alternative methods.

Figure 3. Operational relationships of PPC.

The configuration of user variables and the nature of user input are critical factors
that shape the effectiveness of ML systems, especially in process qualification assessment.
These factors necessitate flexible evaluation criteria that can adapt to different user-defined
settings for learning and prediction intervals, as well as to the complexities introduced by
handling multiple classes for predictions within each category of the qualification process.
Accordingly, the third research question (RQ3) examines how the user interface design of
ML systems can accommodate the diverse requirements introduced by varying user inputs,
such as changes in learning and prediction timeframes, classifications within multi-class
prediction scenarios, and the adjustment of intervention thresholds. This exploration is
vital to ensure that ML models remain robust and accurate when faced with the dynamic
conditions posed by end-user interactions and their individual needs during the process
qualification assessments.
Conclusively, the objective of this study is to examine the capabilities of ML methods
for interactive prediction systems in process quality assessment and compare them to
interpretable models. In addition, it seeks to assess the accuracy of predictive algorithms
for C pk value prediction and compare contemporary prediction algorithms with alternative
methods. Finally, it aims to investigate the alterations in evaluation criteria for ML models
in response to changes in learning and prediction horizons, multi-class prediction per class,
and intervention boundaries.

3. Methodology
Reviewing literature systematically, according to Okoli and Schabram [20], is used as
a methodology for the successive development of relevant information (Figure 4). For this
purpose, the authors define a workflow that explicitly refers to SLA in information sys-
tems research.
Processes 2024, 12, 1730 7 of 20

Records identified through search engines/databases: n = 1398

General Scope: “Process Control”, “Quality Control”

Identification
Detailed Scope: “Machine Learning”, “Prediction”, “Forecast”
Application Area: “Production”, “Manufacturing”

Records excluded after abstract filter (must include: Proc*

After duplicates removed
and/or Qual* - where * is everything after these letters)
n = 1172
n=996

Initial screening Records excluded after abstract screening:

n = 176 n=123
Screening

Subjective filter based on topic related content

of abstract.
Records after screening
n = 53

Full-text papers assessed Records excluded after the full-text screening:

n = 50 n = 10
Eligibility

Subjective filter based on topic related content

of full paper.
Eligible full-text papers
n = 40

Papers identified in reverse search of eligible

Included

Records included in the paper full text papers:

n = 42 n=2

Figure 4. Sequence of the SLA in accordance to Okoli and Schabram [20].

The pre-selected online libraries, namely Google Scholar, Web of Science, and Scopus,
encompass the majority of the literature that is generally accessible, as stated by [21].
However, it cannot be disregarded that relevant literature may have been published beyond
the confines of the selected libraries. Nevertheless, the proportion of literature that is not
typically found in the selected online libraries is relatively small. To ensure that no relevant
works have been overlooked, a reverse search was conducted by examining the references of
the filtered works. This process involves reviewing the cited references within each selected
paper to identify any additional relevant studies that may not have appeared in the initial
search, thereby enhancing the comprehensiveness and accuracy of the literature review.
The search itself is conducted based on various combinations of terms, which are
within the context of PPC. However, prior to this, Vom Brocke et al. [22] highlighted the
impact of selecting predefined search terms on the outcomes. In order to ensure a more
unbiased search, blocks are categorized using keywords and all combinations from these
blocks are inputted as search patterns. This methodology serves to minimize the error
rate in terms of identifying relevant work. For this particular study, the search terms are
divided into three combination blocks. Each block is assigned to a category that holds
significance for the process index. The first block focuses on fundamental topics, such as
“Process Control” and “Quality Control”. This is followed by a secondary determination of
the specific focus of the research. Given the distinct characteristics, two blocks are utilized
for this purpose—one describing the application and the other categorizing the type, thus
thematically classifying the application. The first subordinate block employs the terms
“machine learning”, “prediction”, and “forecast”. The final block completes the search with
the terms “production” and “manufacturing”. Consequently, a total of twelve different
combinations are generated. The search encompasses publications from the past ten years.
The quantitative outcome of the literature search reveals a total of 1398 papers retrieved.
Processes 2024, 12, 1730 8 of 20

4. Results
Due to the research work already carried out on the use of ML in the context of
product and process quality, a broad body of knowledge can be drawn upon. The objective
of the SLA is the elaboration of relevant literature to describe the current findings of PPC.
Furthermore, research gaps related to the dependencies (Figure 2) of PPC are subsequently
identified. Specifically, categories are defined to describe the industrial application of
the research. These target the production engineering background, the applied statistical
methodology, the experimental setup, and the data used.
To emphasize the current relevance and significance of the research endeavor, it
is worth noting that Figure 5 provides a comprehensive illustration of the quantity of
pertinent publications that have emerged in recent years. This visual representation clearly
demonstrates the substantial concentration of scholarly works focusing on the same subject
matter since 2018, which in turn underscores the unwavering commitment and fervor
exhibited by the research community in their pursuit of various inquiries pertaining to PPC.
It is of particular interest to note that a substantial proportion, specifically approximately
65%, of the articles identified in the SLA were disseminated during this specific timeframe,
thereby affirming the heightened level of scholarly discourse.

Figure 5. Periodic development of relevant papers by year and cumulated.

4.1. Industries
Cheng et al. [23] delved into the Catalytic Reforming Processes, which play a pivotal
role in the production of high-octane fuels, thus making a significant stride towards ad-
dressing the conundrum of energy sustainability. Schäfer et al. [24] shed light on the Print-
ing/Painting industry, ambitiously seeking avenues to enhance efficiency and sustainability.
Equally significant is the examination of Casting processes by Mishra and Rane [14], a pro-
cess of paramount importance in ensuring the manufacturing of high-quality components
across various sectors.
In the field of injection molding, which is a crucial component of plastics manufacturing,
significant advancements have been made by researchers such as Ke and Huang [25]. Their
work focuses on enhancing efficiency and promoting sustainability within plastic production
processes. Additionally, the transformative potential of additive manufacturing, particularly in
sectors such as aerospace and healthcare, has been widely explored. Key contributions in this
area have been made by Ismail et al. [26], Krauß et al. [27], and Zhang et al. [28], whose research
is instrumental in advancing the capabilities and applications of 3D printing technologies.
The complex and diverse field of general manufacturing has been thoroughly analyzed by
scholars such as Viharos and Jakab [29], Teinemaa et al. [30], Abbasi [31], and Nalbach et al. [32].
Their research provides valuable insights into a wide range of manufacturing techniques
and technologies, offering significant contributions that impact various industrial sectors.
The semiconductor manufacturing industry, which is critical to the electronics sector, has
also seen groundbreaking advancements through the work of Schrunner et al. [33].
Processes 2024, 12, 1730 9 of 20

In the domain of welding technologies, essential to industries such as construction and

aerospace, important progress has been made by Escobar et al. [34]. Additionally, related
fields have benefited from the research on credit score analysis by Boaventura et al. [35],
which paves the way for more accurate risk evaluation models. Finally, the review high-
lights the contributions of Köksal et al. [36] to molding technologies, particularly in ad-
vanced polymer processing, with implications for the automotive, medical, and consumer
goods industries. Figure 6 presents a visual summary of the industries examined in the
review, highlighting the diverse areas where significant advancements have been made
through research. The figure showcases the breadth of industries impacted by PPC and
relevant studies.

Figure 6. Occurence of different industries found in the relevant literature.

4.2. Data
In the context of PPC, the comparison of datasets from diverse case studies is essential
for understanding the robustness and generalizability of findings. These studies have
been conducted by various researchers and teams, each bringing unique expertise and
perspectives to the analysis. Notably, approximately 19% of the analyzed datasets are
synthetic. Synthetic data hold particular value in scenarios where real-world data are
limited or when specific conditions must be replicated for analysis, offering a flexible and
controlled alternative for testing and validation purposes (Figure 7). The data utilized
in these studies originate from a range of sources, including real-world observations,
controlled experiments, surveys, and synthesized datasets differing in their size (Figure 8).

Figure 7. Proportion of simulated and real data.

Processes 2024, 12, 1730 10 of 20

Figure 8. Distribution of sample size in a logarithmic scale.

The choice of data type depends on research goals and data accessibility. These datasets
have varying sample sizes, based on the research domain and methodology. One challenge
for researchers is ensuring the integrity, dependability, and representativeness of the data.
Data collection, validation, and preprocessing are rigorously emphasized. Additionally,
ethical, privacy, and legal concerns require careful management and safeguarding of
sensitive data. In the context of PPC, the data from case studies are invaluable for scientific
breakthroughs and knowledge in various fields. These data facilitate replication and
validation of research, highlighting the importance of transparency and responsibility
in science.

4.3. Design of Experiments

The studies encompass a broad spectrum of experiments within the realm of manufac-
turing and quality control. For instance, Afrasiab and Khodaygan [37] measured control
points and employed Monte Carlo Simulation to relate them to workpieces, calculating
Capability Ratios for quality assessment. Meanwhile, Schäfer et al. [24] focused on predict-
ing maintenance needs for grippers and failure forecasting systems within an electronics
production cell. On the other hand, Ke and Huang [25] conducted a two-factor full-factorial
experiment to explore correlations between process parameters and quality indicators for
injection molding, using MLP neural networks for prediction.
Several experiments emphasized the importance of ML and predictive modeling.
Li et al. [38] compared ensemble learners with traditional algorithms for predicting steel
quality, while Zan et al. [39] utilized Bi-LSTM to analyze historical data patterns in his-
tograms and line charts. Khoza and Grobler [40] evaluated various ML algorithms for
multivariate SPC. Similarly, Teinemaa et al. [30] and Jeereddy et al. [41] employed ML
techniques for event log evaluation and unit production prediction, respectively.
In addition to quality control and predictive modeling, some experiments concentrated
on optimizing manufacturing processes. Cheng et al. [23] employed an individual adjusted
simple linear regression to improve manufacturing processes, while Wuest et al. [9] used
process state information to adjust parameters during manufacturing, reducing quality
deviations and enhancing product quality.
Furthermore, experiments by Demircioglu Diren et al. [42] and Viharos and Jakab [29]
involved process monitoring and control, using Hotellings T^2 control charts and rein-
forcement learning agents to identify and rectify faults in production processes. Lastly,
various studies emphasized data-driven approaches for quality prediction, such as the
work by Kahya et al. [43], Mabunda and Mashamba [44], and Qiu and Xie [45].
These diverse experiments collectively contribute to advancing quality assurance,
predictive maintenance, and process optimization in the manufacturing sector.
Processes 2024, 12, 1730 11 of 20

4.4. Methodologies
In the field of SPC, researchers from diverse backgrounds have applied ML prin-
ciples to address both classification and regression challenges, employing a variety of
predictive algorithms (Figure 9). For instance, some researchers, such as Afrasiab and
Khodaygan [37], have utilized the Monte Carlo method specifically for regression tasks,
contributing significantly to predictive modeling and forecasting within SPC. Others,
such as Li et al. [38], have tackled regression problems using ensemble learning tech-
niques, including methods such as Linear Regression, Support Vector Machines (SVM),
Random Forests, and Gradient Boosting, to derive comprehensive insights. Additionally,
some studies, exemplified by Abbasi [31], focus exclusively on regression by employing
Artificial Neural Networks (ANNs), a versatile tool in ML. In contrast, Pheng et al. [46]
leveraged LSTM networks, which are particularly well-suited for forecasting tasks.
Moreover, Li et al. [47] combined Convolutional Neural Networks (CNNs) with SVMs
to address regression problems, aiming to enhance predictive accuracy by integrating
the strengths of these algorithms. Further diversifying the approaches, researchers
such as Jeereddy et al. [41] employed a range of algorithms, including XGBoost and
LSTM, to tackle regression challenges across different application domains. These efforts
contribute to the broader advancement of predictive analytics and the refinement of
forecasting models. Finally, Li et al. [47] continued to explore regression methodologies,
seeking to deepen the understanding of predictive capabilities within their specific fields
of study, thereby advancing the field of SPC.

Figure 9. Frequency of predictive algorithms found in the relevant literature.

4.5. Dependent Variables

Within the field of quality control and process monitoring, researchers have examined
a broad spectrum of dependent variables to improve our understanding of industrial pro-
cesses and product quality. This review compiles and analyzes key findings from studies
published between 2008 and 2023, offering a detailed overview of the progress and emerg-
ing trends in this area (Figure 10). Several studies have focused on the geometric aspects of
industrial processes. Afrasiab and Khodaygan [37] examined the importance of six geomet-
rical control points, while Ke and Huang [25] explored the geometric width of products.
These investigations shed light on how geometric factors contribute to product quality
and process control. Quality assessment and control are integral to manufacturing and
production processes. Researchers such as Li et al. [38] and Kahya et al. [43] have examined
product quality through parameters such as yield strength, tensile strength, elongation, and
impact work, providing valuable insights into the factors influencing steel quality. Addi-
tionally, the studies by Schäfer et al. [24] and Mahadevan and Theocharous [48] delved into
negative pressure and inventory levels as indicators of product quality and process stability.
Process monitoring and control are crucial in maintaining efficiency and minimizing defects.
Processes 2024, 12, 1730 12 of 20

Various studies have explored control chart patterns [23,39,45,49], Hotellings T^2 [42], and
abnormal wave profiles [50] to detect deviations and anomalies. These investigations
provide insights into the detection and mitigation of process variations. Furthermore,
researchers have undertaken significant effort of predictive process modeling. Mishra
and Rane [14] investigated the probability of achieving a global benchmark quality level,
while Abbasi [31] delved into the prediction of PCI for distributions. Similarly, the study
by Kabasakal et al. [51] focused on binary predictions for fault detection, highlighting the
importance of reliable modelling structures in quality management. Multiclass and multi-
label classification have also been explored in quality control and process monitoring in
their study. In addition, Demircioglu Diren et al. [42] investigated different types of faults
and critical events, respectively, while Pheng et al. [46] explored multiclass classification
models to enhance process monitoring.

Figure 10. Categories and frequency of dependent variables in the observed literature.

4.6. Model Performance

The performance of the models in these case studies varies, demonstrating the versatility
and complexity of real-world industrial challenges (Figure 11). Some studies have achieved
high accuracy or R-squared values, such as the classification models in Escobar et al. [34]
and Li et al. [52], which both exhibit accuracy scores of 0.99. On the other hand, some regres-
sion models, such as those in Schäfer et al. [24] and Xiao et al. [53], have achieved relatively
high R-squared values of 0.83 and 0.99, showcasing their effectiveness in predicting outcomes.

Figure 11. Comparison of algorithmic performance (Accuracy and R2 ) regarding classification

and regression.
Processes 2024, 12, 1730 13 of 20

It is important to note that the performance of these models often depends on various
factors, including the quality of data preprocessing, feature engineering, and the suitability
of the chosen algorithms for the specific problem at hand. Overall, this dataset provides
valuable insights into the application of data analytics and ML techniques in industrial
settings, showcasing the potential for improved process control, quality monitoring, and
predictive maintenance.

5. Analysis of Results
Due to the extensive investigation of ML algorithms, the existing body of knowledge
can be leveraged for the practical implementation of PPC in industrial settings. However, it
is worth noting that only a limited number of scrutinized algorithms have been employed
for the prediction of Cpk value. Consequently, a combination of interpretable (descriptive)
and advanced (black-box) models is utilized for the purpose of process quality prediction.
Moreover, no experiments have been identified that integrate quality criteria with rein-
forcement learning. As a result, there is room for expanding the scope of this study due to
the scarcity of independent variables and the absence of xAI methods. By examining the
derivations of the SLA and the localization of application areas in PPC, it becomes possible
to identify the pertinent research gaps that exist in the related domains.

5.1. RQ1: Interpretation Domain—Humans

The imperative for human-centered interpretability in manufacturing prediction
systems is becoming increasingly evident as the industry forges ahead in the digital
age [54]. In direct response to this evolution, the research outcomes elucidate the profound
necessity for explainable artificial intelligence (xAI) techniques that stand superior to
classical models, particularly in their ability to elucidate complex ML algorithms for end
users. xAI techniques bridge the cognitive gap between sophisticated ML predictions and
pragmatic decision making by elucidating the underlying logic of numerical, binary, and
multiclass predictions. This cognitive bridge is not merely a convenience but a catalyst
for enhancing productivity and ensuring quality in production processes. It enables
process owners and operators to garner transparent insights and fosters an environment
where informed decisions are the norm rather than the exception. However, only few
case studies show the direct relationship of PPC and interpretable results, as can be seen
in Boaventura et al. [35], Qiu and Xie [45], Kabasakal et al. [51], and Kim et al. [49].
Within digitization, there has been a growing demand for more intuitive and user-
friendly interactive prediction systems in production. As the complexity of manufacturing
processes increases, it becomes essential for end users to have access to accurate, reliable,
and timely information [55]. Given the preceding information, factors that contribute to
the successful implementation of these systems must be explored and requirements need
to be derived for the end user.
First and foremost, the accessibility of the interactive prediction system is paramount.
For end users to effectively harness the benefits of PPC, the system must be easy to under-
stand, navigate, and operate. This means that the user interface should be designed with
simplicity and usability in mind, ensuring that the learning curve is minimal for opera-
tors who may not have advanced technical expertise. In addition to being user-friendly,
the system should also be customizable and adaptable to the specific needs of the produc-
tion environment [56,57]. This involves taking into consideration the unique challenges
and requirements of each manufacturing process, and tailoring the interactive prediction
system accordingly. By offering a high degree of customization, the system becomes more
relevant and valuable to the end user. Moreover, the interactive prediction system should
be equipped with advanced algorithms and artificial intelligence capabilities, enabling it
to learn from historical data and continuously improve its predictions. This self-learning
ability is crucial in adapting to changing production conditions and maintaining a high
level of accuracy and reliability. Furthermore, seamless integration with existing produc-
tion systems and equipment is essential for maximizing the potential of the interactive
Processes 2024, 12, 1730 14 of 20

prediction system. By ensuring compatibility and interoperability with other production

elements, end users can harness the full power of PPC without disrupting their current
operations [58].
In the quest to optimize the performance of interactive prediction systems in production
for end users, interpretability emerges as a crucial factor in ensuring transparency, trust,
and informed decision making. Addressing these requirements calls for a comprehensive
comparison between xAI methods and classical interpretable models in terms of their
ability to provide actionable recommendations to process owners. xAI methods, with their
capacity to unravel the inner workings of complex ML models, exhibit a distinct advantage
over traditional interpretable models by facilitating a more nuanced understanding of the
underlying decision-making process [15]. This advantage enables stakeholders to better
comprehend the rationale behind predictions, ensuring a more informed and context-
sensitive adoption of recommendations. As such, the integration of xAI techniques in
interactive prediction systems is paramount in fostering a transparent and effective decision-
making landscape for end users, ultimately enhancing the practicality and adaptability of
these systems in real-world applications.

5.2. RQ1: Interpretation Domain—Machines

Since PPC relies on the timely acquisition and analysis of sensor data, gathered from
various points throughout the production process, automated decision processes rely
reversely on data quality and stable connections [59]. Sensors detect fluctuations in temper-
ature, pressure, humidity, and other essential factors, providing valuable insight into the
behavior of the system. These data serve as the foundation for building predictive models,
enabling real-time monitoring, and identifying potential issues before they escalate [60].
As more sensors are integrated into the production process, the complexity of managing
this data increases. This poses a significant challenge in creating an environment that can
efficiently process, analyze, and communicate the information. Effective data management
solutions and communication tools are needed to filter, store, and transmit the data to
appropriate channels for analysis and decision making [33,42].
One solution lies in edge computing, a decentralized approach to data processing
that pushes computation to the edge of the network, closer to the source of the data. Edge
computing can significantly reduce the latency and bandwidth required to transmit data to
centralized servers, allowing for real-time analysis and decision making [61]. Implementing
edge computing systems can be complex and costly, especially for small- and medium-sized
enterprises [62].
Another challenge in creating a PPC environment is ensuring seamless communica-
tion between different devices, software, and systems involved in the production process.
The Industrial Internet of Things (IIoT) aims to address this issue by providing a stan-
dardized framework for device-to-device communication [63]. However, interoperability
between devices from different manufacturers remains a challenge, as there is no univer-
sally accepted standard for IIoT communication. Moreover, cybersecurity is an increasingly
critical concern as the use of connected devices and sensors in manufacturing processes
expands. Ensuring the confidentiality, integrity, and availability of production data is
paramount to maintain trust and prevent unauthorized access.

5.3. RQ2: Analysis Domain

Modern algorithms such as LSTMs and Reinforcement Learning have demonstrated
substantial superiority in predictive accuracy, robustness, and model quality when com-
pared to traditional predictive methods [29,39]. The predictive performance of these
algorithms is contingent on several factors, including the underlying process distribution,
noise characteristics, and model complexity [64]. Consequently, it becomes imperative to as-
sess the predictive power of these algorithms across various process distributions, in order
to ascertain their robustness and generalizability to a wide range of industrial applications.
Processes 2024, 12, 1730 15 of 20

These algorithms are designed to facilitate accurate, timely, and efficient decision
making by predicting future behavior and events in a given system, thereby enhancing the
overall performance of process control systems. As industrial processes become increas-
ingly complex and interconnected, the need for dynamic and adaptive control strategies
has never been greater [65]. Consequently, a burgeoning body of research has emerged,
seeking to explore and quantify the advantages offered by modern predictive algorithms in
comparison to traditional control methodologies. The efficacy of these algorithms hinges
upon their ability to construct accurate and reliable predictive models, which can be de-
rived from historical and real-time data, enabling process engineers to implement proactive
adjustments in response to anticipated changes in process conditions [66]. By doing so,
these algorithms can effectively mitigate undesirable fluctuations, reduce the incidence of
disturbances, and optimize operational efficiency.
Central to the success of modern predictive algorithms is their inherent adaptability,
which arises from their ability to learn and evolve in response to changes in process
dynamics, system characteristics, and environmental conditions [67]. Hence, this flexibility
allows these algorithms to be applied across a wide range of processes. Furthermore,
the availability of high-performance computing resources and the proliferation of low-cost,
high-quality sensors have facilitated the generation and processing of massive volumes of
data, thereby enhancing the accuracy and precision of predictive models.
Despite these promising advancements, it is crucial to recognize that the efficacy of
predictive algorithms is contingent upon various factors, including the quality of the input
data, the appropriateness of the selected modeling technique, and the ability to account
for uncertainties and non-linearities in the system. As such, the accurate quantification
of the advantages offered by these algorithms necessitates a comprehensive evaluation
of their performance across a diverse array of process control applications, in order to
establish their relative superiority over traditional control methodologies [68]. Moreover,
it is imperative to consider the potential limitations and challenges associated with the
implementation of predictive algorithms, including the need for robust data preprocess-
ing techniques, the management of computational complexity, and the establishment
of effective feedback mechanisms to ensure the stability and robustness of the control
system [52].
In light of these considerations, it is crucial to systematically investigate the extent
to which modern predictive algorithms enhance process control. Such an investigation is
essential to fully understand the potential of these algorithms across various industrial
applications and to identify opportunities for future research and development in PPC.
By exploring these avenues, industries can effectively harness the power of AI and ML
technologies to drive substantial improvements in the efficiency, safety, and sustainability
of industrial processes, thereby fostering innovation and progress across a broad spectrum
of scientific and engineering domains.

5.4. RQ3: Domain of External Influence

A primary challenge in the interface of ML systems for process qualification assess-
ment is the continuous evolution of user inputs and the associated production conditions.
These conditions necessitate the adaptation of predictive models to reflect user-driven
changes [69]. This challenge calls for the implementation of adaptive ML algorithms ca-
pable of modifying model parameters and the weights of features based on dynamic user
interactions and changes within the production framework. To tackle this issue, researchers
have introduced methods for the ongoing assessment and real-time adjustment of ML
models within live user environments. One prevalent method involves the application
of online learning algorithms that adopt incremental learning practices to refine model
weights with each new piece of user data. Another method is the utilization of a hybrid
learning model, merging offline and online learning to strike an optimal balance between
model consistency and adaptability [70–72].
Processes 2024, 12, 1730 16 of 20

Beyond these technical considerations, the practical deployment of predictive controls

through user interfaces in production settings also presents numerous challenges. A signifi-
cant hurdle is ensuring the availability and reliability of data derived from user interactions,
which, if incomplete or noisy, can adversely affect model predictions. Furthermore, the com-
plexity of industrial systems often hampers the interpretability of models, especially when
the link between user inputs and outcomes is obscured [73]. Overcoming these obstacles
necessitates a dual approach that marries technical innovation with practical solutions.
Additionally, understanding which user inputs or variables have a pivotal role in the
predictive accuracy of ML models in process qualification is an area of active research with
far-reaching consequences for production efficiency [74]. By persistently refining these
models in response to the changing landscape of user engagement, it becomes feasible
to cultivate a more resilient and adaptive system for PPC, thus enabling the handling of
disruptions and promoting optimal production results.

6. Conclusions
This study provides a comprehensive investigation into the application of ML in PPC,
emphasizing its practical implementation across various industrial scenarios. The anal-
ysis reveals the significant potential of ML to enhance accuracy and decision-making
processes in manufacturing. By examining a diverse range of case studies and appli-
cations, the research illustrates the versatility of ML in addressing complex industrial
challenges and optimizing operational strategies [20]. The findings highlight how ML
models, such as supervised learning and neural networks, have been particularly effec-
tive in overcoming the reactive characteristics of traditional SPC, enabling more precise
predictions and informed decision making [14,15]. These models have demonstrated
their capability in synthesizing large volumes of data to predict potential issues before
they occur, thus improving both the efficiency and reliability of manufacturing pro-
cesses [22,59]. For instance, ANNs have shown significant promise in estimating process
capability for non-normal processes, contributing to more accurate and reliable quality
control measures [31]. Additionally, ensemble learning techniques have been effective
in improving the quality control of multistage manufacturing systems by leveraging
data-driven insights for real-time process adjustments [32,73]. Moreover, the integration
of ML into PPC not only enhances predictive accuracy but also facilitates the development
of adaptive control mechanisms that respond dynamically to changes in the manufac-
turing environment. This proactive approach enables manufacturers to anticipate and
mitigate operational challenges, ensuring a more resilient and responsive production pro-
cess [8,27]. Thus, PPC underscores the transformative potential of ML in revolutionizing
manufacturing paradigms by augmenting analytical capabilities and providing real-time
process control. These advancements have contributed to sustained improvements in
process quality and efficiency, marking a significant step forward in the field of industrial
manufacturing [14,28].
However, it is crucial to recognize various limitations inherent in this study. Firstly,
the research may not encompass all possible confounding variables, a restriction that could
incorporate hidden biases and subsequently distort the accuracy of the results. These
variables could arise from overlooked factors during the analysis, such as disparities in
equipment, operator expertise, or prevailing environmental conditions. Addressing these
potential biases requires a more comprehensive approach, integrating a diverse range
of variables to better understand the intricate relationships between ML techniques and
advancements in process quality.
Another significant limitation is the potential absence of reproducibility in the study,
which can have an impact on the level of certainty regarding the resulting outcomes.
The lack of independent replications has the potential to undermine the strength and
generalizability of the results across different industries and settings. This suggests that
the benefits of ML techniques may be specific to certain contexts or that the methods
applied may not be universally applicable across diverse manufacturing environments. It
Processes 2024, 12, 1730 17 of 20

is essential to encourage further research and independent verification of these findings in

order to strengthen the foundational knowledge of ML applications in PPC.
The use of self-reported data in this study introduces biases and inaccuracies due to its
susceptibility to factors such as errors in recall, the desire for social approval, or intentional
misreporting. These biases can distort the perceived impact of ML techniques on process
quality and decision-making processes, either exaggerating or underestimating the actual
influence. Therefore, future research should focus on incorporating more objective and
diverse data sources, such as direct measurements or sensor data, to enhance the reliability
and validity of the findings and provide a more authentic representation of the effectiveness
of ML techniques in PPC.
Addressing these limitations and developing more robust methodologies will be
crucial in establishing conclusive evidence and drawing more reliable conclusions regarding
the effectiveness of ML techniques in PPC, specifically within the steel industry and other
industrial processes. The considerations highlighted in this conclusion aim to promote
a more refined approach to future research efforts in this field, fostering advancements in
industrial process optimization and contributing to the existing body of knowledge on the
intersection of ML and industrial applications.

Author Contributions: Conceptualization, J.M. and R.J.; methodology, J.M.; validation, J.M. and R.J.;
formal analysis, J.M.; investigation, J.M.; resources, J.M.; data curation, J.M.; writing—original draft
preparation, J.M.; writing—review and editing, R.J.; visualization, J.M.; supervision, R.J. All authors
have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: The original contributions presented in the study are included in the
article, further inquiries can be directed to the corresponding author.
Conflicts of Interest: The authors declare that they have no known competing financial interests or
personal relationships that could have appeared to influence the work reported in this paper.

References
1. Goshime, Y.; Kitaw, D.; Jilcha, K. Lean manufacturing as a vehicle for improving productivity and customer satisfaction: A
literature review on metals and engineering industries. Int. J. Lean Six Sigma 2019, 10, 691–714. [CrossRef]
2. Kane, V.E. Process capability indices. J. Qual. Technol. 1986, 18, 41–52. [CrossRef]
3. Tsung, F.; Li, Y.; Jin, M. Statistical process control for multistage manufacturing and service operations: A review and some
extensions. Int. J. Serv. Oper. Inform. 2008, 3, 191–204. [CrossRef]
4. Lee, S.M.; Lee, D.; Kim, Y.S. The quality management ecosystem for predictive maintenance in the Industry 4.0 era. Int. J. Qual.
Innov. 2019, 5, 4. [CrossRef]
5. Zhang, Y.; Zhang, G.; Wang, J.; Sun, S.; Si, S.; Yang, T. Real-time information capturing and integration framework of the internet
of manufacturing things. Int. J. Comput. Integr. Manuf. 2015, 28, 811–822. [CrossRef]
6. Das, A.; Mondal, S.; Thakkar, J.; Maiti, J. A methodology for modeling and monitoring of centrifugal casting process. Int. J. Qual.
Reliab. Manag. 2015, 32, 718–735. [CrossRef]
7. Dombrowski, U.; Wullbrandt, J.; Krenkel, P. Industrie 4.0 in production ramp-up management. Procedia Manuf. 2018, 17, 1015–1022.
[CrossRef]
8. Guh, R.S. Integrating artificial intelligence into on-line statistical process control. Qual. Reliab. Eng. Int. 2003, 19, 1–20. [CrossRef]
9. Wuest, T.; Irgens, C.; Thoben, K.D. An approach to monitoring quality in manufacturing using supervised machine learning on
product state data. J. Intell. Manuf. 2014, 25, 1167–1180. [CrossRef]
10. Lee, S.; Amin, R.W. Process tolerance limits. Total Qual. Manag. 2000, 11, 267–280. [CrossRef]
11. Sharma, G.; Rao, P.S. A DMAIC approach for process capability improvement an engine crankshaft manufacturing process. J. Ind.
Eng. Int. 2014, 10, 65. [CrossRef]
12. Dietrich, E.; Conrad, S. Statistische Verfahren zur Maschinen-und Prozessqualifikation; Carl Hanser Verlag GmbH Co. KG: München,
Germany, 2021; ISBN 978-3-446-46447-6.
13. Hastie, T.; Tibshirani, R.; Friedman, J.; Hastie, T.; Tibshirani, R.; Friedman, J. Unsupervised learning. In The Elements of Statistical
Learning; Springer: New York, NY, USA, 2009. [CrossRef]
14. Mishra, N.; Rane, S.B. Prediction and improvement of iron casting quality through analytics and Six Sigma approach. Int. J. Lean
Six Sigma 2019, 10, 189–210. [CrossRef]
15. Senoner, J.; Netland, T.; Feuerriegel, S. Using explainable artificial intelligence to improve process quality: Evidence from
semiconductor manufacturing. Manag. Sci. 2022, 68, 5704–5723. [CrossRef]
Processes 2024, 12, 1730 18 of 20

16. Rehse, J.R.; Mehdiyev, N.; Fettke, P. Towards explainable process predictions for industry 4.0 in the dfki-smart-lego-factory.
KI-Künstliche Intell. 2019, 33, 181–187. [CrossRef]
17. Rudin, C.; Radin, J. Why are we using black box models in AI when we do not need to? A lesson from an explainable AI
competition. Harv. Data Sci. Rev. 2019, 1, 1–9. [CrossRef]
18. Hyun Park, S.; Seon Shin, W.; Hyun Park, Y.; Lee, Y. Building a new culture for quality management in the era of the Fourth
Industrial Revolution. Total Qual. Manag. Bus. Excell. 2017, 28, 934–945. [CrossRef]
19. Sharma, M.; Luthra, S.; Joshi, S.; Kumar, A. Implementing challenges of artificial intelligence: Evidence from public manufacturing
sector of an emerging economy. Gov. Inf. Q. 2022, 39, 101624. [CrossRef]
20. Okoli, C.; Schabram, K. A guide to conducting a systematic literature review of information systems research. Commun. Assoc.
Inf. Syst. 2015, 37, 879–910. [CrossRef]
21. Gusenbauer, M. Google Scholar to overshadow them all? Comparing the sizes of 12 aca-demic search engines and bibliographic
databases. Scientometrics 2019, 118, 177–214. [CrossRef]
22. Vom Brocke, J.; Simons, A.; Riemer, K.; Niehaves, B.; Plattfaut, R.; Cleven, A. Standing on the shoulders of giants: Challenges and
recommendations of literature search in information systems research. Commun. Assoc. Inf. Syst. 2015, 37, 9. [CrossRef]
23. Cheng, W.K.; Azman, A.F.; Hamdan, M.H.; Mansa, R.F. Application of Six Sigma in oil and gas industry: Converting operation
data into business value for process prediction and quality control. In Proceedings of the 2014 IEEE International Conference on
Industrial Engineering and Engineering Management, Selangor, Malaysia, 9–12 December 2014; IEEE: Piscataway, NJ, USA, 2014;
pp. 148–153. [CrossRef]
24. Schäfer, F.; Schwulera, E.; Otten, H.; Franke, J. From descriptive to predictive six sigma: Machine learning for predictive
maintenance. In Proceedings of the 2019 Second International Conference on Artificial Intelligence for Industries (AI4I), Laguna
Hills, CA, USA, 25–27 September 2019; IEEE: Piscataway, NJ, USA, 2019; pp. 35–38. [CrossRef]
25. Ke, K.C.; Huang, M.S. Quality classification of injection-molded components by using quality indices, grading, and machine
learning. Polymers 2021, 13, 353. [CrossRef] [PubMed]
26. Ismail, M.; Mostafa, N.A.; El-assal, A. Quality monitoring in multistage manufacturing systems by using machine learning
techniques. J. Intell. Manuf. 2021, 33, 2471–2486. [CrossRef]
27. Krauß, J.; Pacheco, B.M.; Zang, H.M.; Schmitt, R.H. Automated machine learning for predictive quality in production. Procedia
CIRP 2020, 93, 443–448. [CrossRef]
28. Zhang, S.; Gao, J.G.; Wang, H.; Zhang, X.Q. Position Tolerance Process Capability Prediction under Maximum Material Condition.
Adv. Mater. Res. 2014, 945, 107–110. [CrossRef]
29. Viharos, Z.J.; Jakab, R. Reinforcement learning for statistical process control in manufacturing. Measurement 2021, 182, 109616.
[CrossRef]
30. Teinemaa, I.; Dumas, M.; Rosa, M.L.; Maggi, F.M. Outcome-oriented predictive process monitoring: Review and benchmark.
ACM Trans. Knowl. Discov. Data (TKDD) 2019, 13, 1–57. [CrossRef]
31. Abbasi, B. A neural network applied to estimate process capability of non-normal processes. Expert Syst. Appl. 2009, 36, 3093–3100.
[CrossRef]
32. Nalbach, O.; Linn, C.; Derouet, M.; Werth, D. Predictive quality: Towards a new understanding of quality assurance using
machine learning tools. In Proceedings of the Business Information Systems: 21st International Conference, BIS 2018, Berlin,
Germany, 18–20 July 2018; pp. 30–42. [CrossRef]
33. Schrunner, S.; Scheiber, M.; Jenul, A.; Zernig, A.; Kästner, A.; Kern, R. Towards a General Framework to Embed Advanced
Machine Learning in Process Control Systems. arXiv 2021, arXiv:2103.13058. [CrossRef].
34. Escobar, C.A.; Morales-Menendez, R.; Macias, D. Process-monitoring-for-quality—A machine learning-based modeling for rare
event detection. Array 2020, 7, 100034. [CrossRef]
35. Boaventura, L.L.; Ferreira, P.H.; Fiaccone, R.L. On flexible statistical process control with artificial intelligence: Classification
control charts. Expert Syst. Appl. 2022, 194, 116492. [CrossRef]
36. Köksal, G.; Batmaz, I.; Testik, M.C. A review of data mining applications for quality improvement in manufacturing industry.
Expert Syst. Appl. 2011, 38, 13448–13467. [CrossRef]
37. Afrasiab, H.; Khodaygan, S. Capability Prediction of Machining Processes Based on Uncertainty Analysis. Int. J. Mech. Mechatron.
Eng. 2016, 10, 1262–1269. [CrossRef]
38. Li, F.; Wu, J.; Dong, F.; Lin, J.; Sun, G.; Chen, H.; Shen, J. Ensemble machine learning systems for the estimation of steel quality
control. In Proceedings of the 2018 IEEE International Conference on Big Data (Big Data), Seattle, WA, USA, 10–13 December
2018; IEEE: Piscataway, NJ, USA, 2018; pp. 2245–2252. [CrossRef]
39. Zan, T.; Liu, Z.; Su, Z.; Wang, M.; Gao, X.; Chen, D. Statistical process control with intelligence based on the deep learning model.
Appl. Sci. 2019, 10, 308. [CrossRef]
40. Khoza, S.C.; Grobler, J. Comparing machine learning and statistical process control for predicting manufacturing performance. In
EPIA Conference on Artificial Intelligence; Springer International Publishing: Cham, Switzerland, 2019. [CrossRef]
41. Jeereddy, S.; Kennedy, K.; Duffy, E.; Walker, A.; Vorster, B. Machine learning use cases for smart manufacturing kpis. In
Proceedings of the 2019 IEEE International Conference on Big Data (Big Data), Los Angeles, CA, USA, 9–12 December 2019; IEEE:
Piscataway, NJ, USA, 2019; pp. 4375–4380. [CrossRef]
Processes 2024, 12, 1730 19 of 20

42. Demircioglu Diren, D.; Boran, S.; Cil, I. Integration of machine learning techniques and control charts in multivariate processes.
Sci. Iran. 2020, 27, 3233–3241. [CrossRef]
43. Kahya, A.S.; Şişmanoğlu, S.; Erçin, Z.; Akdağ, H.C. Total Quality Management through Defect Detection in Manufacturing
Processes Using Machine Learning Algorithms. In Proceedings of the International Symposium for Production Research 2019; Springer
International Publishing: Cham, Switzerland, 2020. [CrossRef]
44. Mabunda, M.R.; Mashamba, A. Predictive Process Control Framework for Online Quality Control in a Hot Rolling Mill. J. Hum.
Earth Future 2022, 3, 263–279. [CrossRef]
45. Qiu, P.; Xie, X. Transparent sequential learning for statistical process control of serially correlated data. Technometrics 2022, 64,
487–501. [CrossRef]
46. Pheng, T.; Chuluunsaikhan, T.; Ryu, G.; Kim, S.H.; Nasridinov, A.; Yoo, K.H. Prediction of process quality performance using
statistical analysis and long short-term memory. Appl. Sci. 2022, 12, 735. [CrossRef]
47. Li, B.H.; Zhao, L.P.; Yao, Y.Y. Multiconditional machining process quality prediction using deep transfer learning network. Adv.
Manuf. 2023, 11, 329–341. [CrossRef]
48. Mahadevan, S.; Theocharous, G. Optimizing production manufacturing using reinforcement learning. In Proceedings of the
FLAIRS Conference, Sanibel Island, FL, USA, 18–20 May 1998; Volume 372, p. 377, ISBN 978-1-57735-051-4.
49. Kim, S.B.; Jitpitaklert, W.; Park, S.K.; Hwang, S.J. Data mining model-based control charts for multivariate and autocorrelated
processes. Expert Syst. Appl. 2012, 39, 2073–2081. [CrossRef]
50. Chou, S.H.; Chang, S.; Tsai, T.R.; Lin, D.K.; Xia, Y.; Lin, Y.S. Implementation of statistical process control framework with machine
learning on waveform profiles with no gold standard reference. Comput. Ind. Eng. 2020, 142, 106325. [CrossRef]
51. Kabasakal, İ.; Keskin, F.D.; Koçak, A.; Soyuer, H. A prediction model for fault detection in molding process based on logistic
regression technique. In Proceedings of the International Symposium for Production Research 2019, Vienna, Austria, 28–30
August 2019; Springer International Publishing: Cham, Switzerland, 2020; pp. 351–360. [CrossRef]
52. Li, J.; Liu, L.; Le, T.D.; Liu, J. Accurate data-driven prediction does not mean high reproducibility. Nat. Mach. Intell. 2020, 2, 13–15.
[CrossRef]
53. Xiao, X.; Waddell, C.; Hamilton, C.; Xiao, H. Quality prediction and control in wire arc additive manufacturing via novel machine
learning framework. Micromachines 2022, 13, 137. [CrossRef] [PubMed]
54. Crankshaw, D.; Gonzalez, J. Prediction-Serving Systems: What happens when we wish to actually deploy a machine learning
model to production? Queue 2018, 16, 83–97. [CrossRef]
55. Denkena, B.; Dittrich, M.A.; Wilmsmeier, S. Automated production data feedback for adaptive work planning and production
control. Procedia Manuf. 2019, 28, 18–23. [CrossRef]
56. Ziegler, D.; Peissner, M. Enabling Accessibility through Model-Based User Interface Development. In Proceedings of the AAATE
Conference, Sheffield , UK, 11–15 September 2017; pp. 1067–1074. [CrossRef]
57. Knowles, R.; Sanchez-Torron, M.; Koehn, P. A user study of neural interactive translation prediction. Mach. Transl. 2019,
33, 135–154. [CrossRef]
58. Gupta, P.K.; Tyagi, V.; Singh, S.K. Predictive Computing and Information Security; Springer: Singapore, 2017. [CrossRef]
59. Sadati, N.; Chinnam, R.B.; Nezhad, M.Z. Observational data-driven modeling and optimization of manufacturing processes.
Expert Syst. Appl. 2018, 93, 456–464. [CrossRef]
60. Chen, W.; Liu, H.; Qi, E. Discrete event-driven model predictive control for real-time work-in-process optimization in serial
production systems. J. Manuf. Syst. 2020, 55, 132–142. [CrossRef]
61. Tulkinbekov, K.; Kim, D.H. Blockchain-enabled approach for big data processing in edge computing. IEEE Internet Things J. 2022,
9, 18473–18486. [CrossRef]
62. Ganapathy, A. Edge computing: Utilization of the internet of things for time-sensitive data processing. Asian Bus. Rev. 2021,
11, 59–66. [CrossRef]
63. Qi, Q.; Xu, Z.; Rani, P. Big data analytics challenges to implementing the intelligent Industrial Internet of Things (IIoT) systems in
sustainable manufacturing operations. Technol. Forecast. Soc. Chang. 2023, 190, 122401. [CrossRef]
64. Rajagopal, R.; Del Castillo, E.; Peterson, J.J. Model and distribution-robust process optimization with noise factors. J. Qual.
Technol. 2005, 37, 210–222. [CrossRef]
65. Vilalta, R.; Apte, C.V.; Hellerstein, J.L.; Ma, S.; Weiss, S.M. Predictive algorithms in the management of computer systems. IBM
Syst. J. 2002, 41, 461–474. [CrossRef]
66. Ławryńczuk, M. Explicit nonlinear predictive control algorithms with neural approximation. Neurocomputing 2014, 129, 570–584.
[CrossRef]
67. Warriach, E.U.; Ozcelebi, T.; Lukkien, J.J. A comparison of predictive algorithms for failure prevention in smart environment
applications. In Proceedings of the 2015 International Conference on Intelligent Environments, Prague, Czech Republic, 15–17
July 2015; IEEE: Piscataway, NJ, USA, 2015; pp. 33–40. [CrossRef]
68. Malyuta, D.; Açikmeşe, B.; Cacan, M. Robust model predictive control for linear systems with state and input dependent
uncertainties. In Proceedings of the 2019 American Control Conference (ACC), Philadelphia, PA, USA, 10–12 July 2019; IEEE:
Piscataway, NJ, USA, 2019; pp. 1145–1151. [CrossRef]
69. Angione, C.; Lió, P. Predictive analytics of environmental adaptability in multi-omic network models. Sci. Rep. 2015, 5, 15147.
[CrossRef]
Processes 2024, 12, 1730 20 of 20

70. Muhr, D.; Tripathi, S.; Jodlbauer, H. An adaptive machine learning methodology to determine manufacturing process parameters
for each part. Procedia Comput. Sci. 2021, 180, 764–771. [CrossRef]
71. Wu, Z.; Rincon, D.; Christofides, P.D. Real-time adaptive machine-learning-based predictive control of nonlinear processes. Ind.
Eng. Chem. Res. 2019, 59, 2275–2290. [CrossRef]
72. Shahbazi, Z.; Byun, Y.C. Smart manufacturing real-time analysis based on blockchain and machine learning approaches. Appl.
Sci. 2021, 11, 3535. [CrossRef]
73. Kozłowski, E.; Mazurkiewicz, D.; Żabiński, T.; Prucnal, S.; S˛ep, J. Machining sensor data management for operation-level
predictive model. Expert Syst. Appl. 2020, 159, 113600. [CrossRef]
74. Moldovan, D.; Cioara, T.; Anghel, I.; Salomie, I. Machine learning for sensor-based manufacturing processes. In Proceedings
of the 2017 13th IEEE International Conference on Intelligent Computer Communication and Processing (ICCP), Cluj-Napoca,
Romania, 7–9 September 2017; IEEE: Piscataway, NJ, USA, 2017; pp. 147–154. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.