Measurement 217 (2023) 113051

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

A novel approach to using artificial intelligence in coordinate metrology

including nano scale
Michal Wieczorowski a, Dawid Kucharski a, Pawel Sniatala b, Pawel Pawlus c,
Grzegorz Krolczyk d, *, Bartosz Gapinski a
a
Institute of Mechanical Technology, Faculty of Mechanical Engineering, Poznan University of Technology, Piotrowo 3, 60-965 Poznan, Poland
b
Faculty of Computing and Telecommunications, Poznan University of Technology, Piotrowo 3, 60-965 Poznan, Poland
c
Department of Manufacturing Processes and Production Engineering, Faculty of Mechanical Engineering and Aeronautics, Rzeszow University of Technology,
Powstancow Warszawy 8, 35-959 Rzeszow, Poland
d
Faculty of Mechanical Engineering Opole University of Technology, Proszkowska 76, 45-758 Opole, Poland

A R T I C L E I N F O A B S T R A C T

Keywords: The applications of modern length and angle metrology continue to pose new challenges for manufacturers and
Artificial intelligence operators of measuring equipment, starting from nano scale. One method to support measurement processes is
Metrology the use of artificial intelligence. As an essential part of Industry 4.0 and Metrology 4.0 strategies, it actively
Scale
supports humans in executing measurement activities. The use of artificial intelligence in this area is determined
Scanning
Nano
by technological and social factors, as the shortage of highly skilled operators is becoming increasingly acute.
The paper discusses the possibilities of using artificial intelligence in coordinate metrology. Ideas including the
selection of the measurement strategy, conditions, filtering techniques, as well as self-verification or even self-
calibration were presented. Schemes of procedures using local and global databases of applications and
measuring instruments are presented. A new area application in the analysis of interference fringes is shown,
with method descriptions and experimental results.

1. Intoduction equipment. The growing electronic, optical, and biomedical industries

are forcing new solutions in the metrology of geometric dimensions.
We live in an era of computerization in almost every area of life. The They also cause the emergence of new concepts related to mechanical
industry has entered a time of digitization and the Internet of Things. As components since it is impossible to have devices and systems that
a result, manufactured devices are becoming smaller and, at the same perform complex execution tasks without them. Metrology is a science
time, more precisely made. Metrology plays a vital role in this trans­ of measurement methods and interpretation principles of the results
formation process. Manufacturing and testing have become more ac­ obtained that coexist with other fields of science very closely, allowing a
curate because of the development of more accurate measurement numerical assessment of phenomena. With this, metrology is also
systems. Advances in electronics and the design of high-speed optical experiencing a dynamic development [2]. Factors that force this and
detectors and stable lasers have contributed to the growing importance have a definite impact include the use of new principles of physics in
of optical metrology. Optical techniques gained further importance after practice, the development of new computer techniques (including
the redefinition of SI units, where the current unit of a meter is defined emerging quantum computers as an opportunity to make the next in­
by the constant c (the speed of light in a vacuum) and the time deter­ dustrial revolution), as well as new manufacturing methods (including
mined using atomic clocks [1]. A coherent light source, a laser with a additive manufacturing) and progressive automation and robotization,
defined wavelength, has become the precise “ruler” in optical length aimed at eliminating man as the most unreliable element of the pro­
measurements. duction process. Not insignificant here are also the Industry 4.0 strategy
The applications of today’s length and angle metrology continue to and its metrological counterpart, Metrology 4.0, which has changed the
pose new challenges for manufacturers and operators of measuring approach to many elements in industry and the perception of everyday

* Corresponding author at: Faculty of Mechanical Engineering Opole University of Technology, Proszkowska 76, 45-758 Opole, Poland.
E-mail addresses: michal.wieczorowski@put.poznan.pl (M. Wieczorowski), dawid.kucharski@put.poznan.pl (D. Kucharski), pawel.sniatala@put.poznan.pl
(P. Sniatala), ppawlus@prz.edu.pl (P. Pawlus), g.krolczyk@po.edu.pl (G. Krolczyk), bartosz.gapinski@put.poznan.pl (B. Gapinski).

https://doi.org/10.1016/j.measurement.2023.113051
Received 12 November 2022; Received in revised form 10 May 2023; Accepted 15 May 2023
Available online 18 May 2023
0263-2241/© 2023 Elsevier Ltd. All rights reserved.
M. Wieczorowski et al. Measurement 217 (2023) 113051

life. All these elements affect the possibility of constructing new and such kind of work was not published before for coordinate metrology,
improved measuring instruments. These instruments, apart from there is definitely a need to show possibilities, probable applications and
advanced mechanical and electronic solutions for reliable measurement future trends.
results, require also elaborating a proper calibration procedure, already
existing for systems commonly used so far. They also need increasingly 2. Artificial intelligence
sophisticated IT solutions (where IT stands for Information Technology
and refers to the use of technology to manage, store, process, and Artificial intelligence denoted as AI refers to the simulation of human
transmit information) based on the right hardware components and intelligence in machines that are programmed to think and learn like
software to assist and guide the operator in making difficult decisions. humans. It has significantly impacted various scientific fields, and
Concepts of length and angle metrology in four dimensions based on spheres of everybody’s life since the first concepts on the subject were
monitoring the dynamic variation of the results over time are also developed. Increasingly, it is becoming apparent that at least its essen­
emerging. tial elements are indispensable in the world around us, and, although
In recent years, geometric quantities’ metrology has been based on often unconsciously, we are eager to take advantage of the benefits it
the coordinate measurement technique. Regardless of the scale at which brings. The area related to measurement devices and measurement itself
measurements are implemented (nano, micro, meso, or macro), the (I&M) is one example where the field of application is still growing. The
object under study is represented as a point cloud from which geometric number of publications on the subject is also growing, indicating the
features or parameters can be determined. This is particularly important scientific community’s great interest. At the same time, it can be
in industrial applications because by ensuring correct dimensions, assumed that the application of AI in I&M is expected to simplify the
components can work together, shapes conform to specifications, and entire measurement procedure from the operator’s point of view and
surface irregularities achieve the functional effect intended by the contribute to improving the accuracy parameters of the measurement
designer. This also makes the equipment provide the requirements, method or measuring instrument [5]. AI methods enable classification,
forcing the people who manage and operate them to be increasingly clustering, prediction, decision-making, and optimization based on
knowledgeable and skilled. However, since such an operator is slowly collected data rather than on a mostly a priori analytical model of the
becoming a scarce and relatively expensive commodity from the point of problem defined based on assumptions. The class of AI methods is in­
view of managing the manufacturing process, a substitute may be found clusive. It includes machine learning (ML), expert systems, Bayesian
in support procedures based on artificial intelligence. This rapidly networks, symbolic intelligence, fuzzy logic and rough set, evolutionary
increasing cost of employing a worker and his time-consuming complete computing and many others. Machine learning (ML) is the most popular
preparation for work on modern instruments favor a work based on and widely used AI method. The classification of activities based on
artificial intelligence in the metrology of geometric quantities. It should machine learning is shown in Fig. 1.
also be noted that during training, an unprepared and poorly trained The first level of the classification includes Neural Networks, Deci­
worker can make a mistake with significant consequences due to inex­ sion Tree, Support Vector Machine and Reinforcement Learning(RL). RL
perience and weak knowledge of the various activities. Moreover, the is a type of machine learning technique that enables an agent to learn in
increasing number of options offered by measurement software manu­ an interactive environment by trial and error using feedback from its
facturers strongly increases its capabilities and may confuse the choice. own actions and experiences. Reinforcement learning is considered its
According to psychological knowledge, too many options to choose from own branch of machine learning, though it does have some similarities
are a source of stress for humans and can cause paralysis and dissatis­ to other types of machine learning. Neural Networks, presented in Fig. 1,
faction [3]. Therefore, any hint from the information system regarding can be divided into Shallow Neural Networks and Deep Neural Net­
this choice can significantly help the work. It is also a sure solution if the works. The terms “shallow” and “deep” refer to the number of layers in a
measurement system developers propose default options based on neural network. Shallow neural networks refer to a neural network that
experience and applications. Developing this thought, we come to ap­ has a small number of layers, usually regarded as having a single hidden
plications based on augmented or virtual reality and artificial intelli­ layer, and deep neural networks refer to neural networks that have
gence, which can be helpful, especially in decision-making processes. multiple hidden layers. Reinforcement learning is considered its own
Thanks to such solutions, it will become possible to make decisions branch of machine learning, though it does have some similarities to
efficiently and quickly based on the data obtained and the information other types of machine learning. Conceptually, we can assume that all
from the knowledge base created and developed based on previous machine learning is a category of algorithms that focus on finding spe­
tasks. cific patterns in the data and using them for prediction. It can be
This review article describes the possibilities of using artificial in­ formalized naturally by defining the objective function Y = F(X),
telligence in various aspects in coordinate metrology. Support for the creating a mapping and dependence of the output variable Y on the
operator, ensuring the correct functioning of measurement systems with input variable X. Machine learning can also be classified using differ­
the problems of the labor market with qualified personnel is naturally a ences in input preparation. Here we can find:
very important element here. AI has also the potential to significantly - Supervised learning, occurring when an input sample labeled ‘i’ has
improve the performance of coordinate metrology by analyzing large both an input feature and a corresponding output label yi.
amounts of data, performing feature extraction, creating models to - Unsupervised learning occurs when an input sample labeled as ‘i’
predict the behavior of an object, performing self-diagnostics and self- has only an input feature and no corresponding output label yi. In this
calibration, and enabling remote monitoring and control capabilities. view, the primary goal of AI operation is to classify all samples in their
These capabilities can lead to more efficient and effective measurement space. This is achieved using techniques such as cluster analysis.
systems that are able to adapt to changing conditions and provide ac­ - Reinforcement learning is based on promoting desirable behaviors
curate and reliable measurements. Such an approach is a novelty from and/or reacting negatively to undesirable phenomena. In this scheme of
the point of view of the metrological approach: instruments with the operation, a reinforcement learning agent can perceive and interpret its
described capabilities are so far at the best of times in the design phase, environment, take action, and learn through trial and error.
and thought is also only being given to creating applied databases from Artificial intelligence methods related to metrology can be helpful
which AI could learn to predict phenomena. Many of the functions primarily for detection, tracking, monitoring, characterization, identi­
described below are not yet present in current measuring devices, but fication, estimation, diagnosis, or prediction. In such an application, as
the presented possibilities show that the application of artificial intel­ early as 1998, the use of neural networks, fuzzy logic, and expert sys­
ligence methods in metrology will soon become necessary [4]. This work tems was described [6]. Thus, the described activity can be used in
is not intended to be a review of AI applications in metrology. Still, as sensor design, calibration, uncertainty prediction, interpolation of

2
M. Wieczorowski et al. Measurement 217 (2023) 113051

Fig. 1. Classification tree of machine learning methods.

measurement data, software elaboration, as well as in indirect mea­ algorithms. They also pointed out the need for optimization during the
surement, fault detection, and system identification [7]. Typically, growing amount of data, among other things, by data compression. Deep
modeling measurement systems at a high level of accuracy of the rep­ learning was also used by Liu et al. for surface defect detection using
resentation of reality are complex and, at the same time, nonlinear and light scattering [18]. Machine learning also includes case-based
dynamic. As a result, defining such systems is impossible, creating space reasoning as a method by which experience is used to solve current
for artificial intelligence. An example of such a situation presented in the problems [19]. This technique has already found application in
literature is a vision-based measurement (VBM) [8,9]. In the publication metrology. The authors of the paper [20] used measurements from the
[10] the authors presented an application to measure the number of production line to correlate dimensions and settings. As seen from this
calories in food based on a photo of the food analyzed. brief overview, the field of application of artificial intelligence from a
Another example related to metrology is the use of AI in the cali­ metrology point of view is growing steadily and dynamically. It is also
bration process, presented by Zhou et al. [11]. This paper uses a photo of impossible not to mention here the Industry 4.0 and Metrology 4.0
a user’s hand to calibrate a relatively complex force measurement device strategies, which further open up new application possibilities for this
easily. Networks of this type are widely used in speech and computer rapidly growing field of computer science. As an example, dealing with
vision systems [12]. Another deep learning method, recurrent neural big data will not be possible without AI procedures, that will be able to
networks (RNNs), are widely used in verbal and non-verbal communi­ show the operator only the most important data, to draw proper
cation activities, particularly in speech recognition, language modeling, conclusions.
translation, and image captioning [13]. Long Short-Term Memory
(LSTM) networks are a type of recurrent neural network capable of 3. Artificial intelligence for coordinate measurements at various
learning order dependence in sequence prediction problems. LSTM units scales
include a ’memory cell’ that can maintain information in memory for
long periods of time. A set of gates is used to control when information Coordinate metrology is a branch of metrology that deals with the
enters the memory, when it’s output, and when it’s forgotten. This ar­ measurement of the geometric characteristics of objects, such as their
chitecture lets them learn longer-term dependencies. It can be used in shape, size, and position. AI can be used to improve the performance of
complex problem domains like machine translation, speech recognition, coordinate metrology by analyzing large amounts of data and identi­
and more. A similar issue is implementing another kind of deep learning: fying patterns that can be used to improve the accuracy and precision of
generative adversarial networks (GAN) work well in game development measurements. Metrology related to the measurement of length and
and the generation of artificial videos [14], and Boltzmann deep ma­ angle operates at all scales, from nano- and microscale to macro-scale,
chines find their application in object recognition. Autoencoding, on the where distances might be so large that there are ideas for using drones
other hand, is very helpful in compressing information and can identify to move measuring devices [21]. Therefore, artificial intelligence ap­
the fault of rotating machinery and analyze the state of the monitored plies to a wide range of diverse applications. Therefore, the macroscale
system. Examples of a broader view of the use of artificial intelligence in mentioned above, considered significant, definitely refers to the mea­
science and technology have also already appeared in the reference surement of geometric features, whereas the microscale or even the
[15]. Heizmann et al. presented the application of artificial intelligence nanoscale covers minor features in terms of dimensions, i.e., asperities
with machine learning in optical measurement and inspection systems occurring on the surface [22]. The area between these two scales is
[16]. They also described the challenges the technique faces in practical mesoscale or intermediate scale. Its domain is the range where the
applications, such as the increasing demand for computer processing topography of the surface and the asperities found on it begin to trans­
power or procedures for estimating uncertainty in ML measurements. form into geometric elements or features for which geometric di­
Genty et al. [17] proposed an exciting aspect of machine learning in mensions are more critical. Examples include high-resolution laser
ultra-fast photonics applications. They comprehensively compared rulers [23].
tuning methods for ultrafast fiber lasers based on machine learning The potential uses of artificial intelligence in the coordinate

3
M. Wieczorowski et al. Measurement 217 (2023) 113051

metrology related to geometric quantities are wide-ranging. AI can be when using simulation methods based on virtual measurement devices
used in coordinate metrology is through the development of computer [24,25]. It will be possible to assess the quality of the obtained mea­
vision algorithms that can be used to analyze images of objects and surement result (based on the estimated measurement uncertainty). If
extract geometric information. These algorithms can perform feature the result is unsatisfactory, the system will suggest alternative actions to
extraction, as the process of identifying and extracting specific features obtain more reliable results. Furthermore, the system can more
of an object, such as its edges, corners, or surface patterns. This infor­ comprehensively learn the limitations of specific measuring instruments
mation can then be used to create a 3D model of the object, for further at the user’s site based on realized applications and develop the local
analysis and measurements. Data from e.g. machine learning algorithms base. Finally, it can predict future applications based on those already
can be utilized to analyze data from sensors and predict the behavior of realized in a given laboratory. Moreover, while we are on the subject of
an object under different conditions or predict the shape and position of simulations, AI can also improve the measurement process based on a
an object, which can be used to improve the accuracy and precision of previous simulation based on Digital Twin.
measurements. An important element of measuring instruments operating based on
An important issue is also related to operator support and simplifi­ Metrology 4.0 using artificial intelligence will be the connectivity of
cation of measurement activities from the operator’s point of view. For a instrument components with the outside world based on the Internet of
measurement technique selection, one can find an application field for Things (IoT). This will allow the measuring instrument to connect, e.g.,
artificial intelligence, as shown in Fig. 2. with a machine tool, from which it can read, among other things, offset
The system operates on two types of databases: local and global. data files and machining process data, based on which AI will suggest
When a measurement application appears, it first searches for an iden­ measurement conditions and settings. It can also warn that obtaining the
tical or similar one in the local database (for a particular company or correct measurement results with specific machine tool settings will be
enterprise). If one cannot be found, the system searches in the global impossible. Furthermore, while we are on the connectivity of the mea­
database managed by an independent entity or measurement equipment surement system with production equipment, it is worth mentioning
manufacturer. Based on the case found, the system determines the Product Manufacturing Information (PMI) as a set containing not only
measurement technique and checks whether a device that functions on geometric features and tolerances but also material data, surface re­
its basis is available at the plant. If not, it proposes selecting a technique quirements, measurement conditions, etc., contained in the 3D model of
that meets the requirements or a specific measuring device, depending the workpiece. This way, instead of relying on digital documents to
on who manages the global base (it should be a manufacturer- convey engineering or manufacturing data, it can all be embedded
independent unit). It may also suggest combining different measure­ within the original CAD file. Taking into account AI, the concept of
ment techniques, such as optical and contact. The final result is the se­ product manufacturing information seems to be the most critical infor­
lection of measurement conditions and the measurement itself. AI can mation in the digital product model [26]. Based on this information
also actively participate in this process. Based on calibration data, it can obtained from CAD data, it will be possible to select the measurement
learn which areas of the measurement volume are more accurate and technique for the application using artificial intelligence.
suggest positioning the object to meet the requirements of the specifi­ Moving on to scale-dependent specific capabilities and starting at the
cations better. macro scale, the measuring devices found here are designed in the
AI is also essential in estimating measurement uncertainty, especially convention of the coordinate measurement technique. This includes

Fig. 2. Artificial intelligence in the selection process of measurement technology.

4
M. Wieczorowski et al. Measurement 217 (2023) 113051

figures and solids generated by the software based on the coordinates to the specifics of these measurement systems. Thus, for a scanner, the
collected from the measuring points and the free surfaces characterized important elements to be supported by artificial intelligence are the
by these points [27]. Among the most widely used coordinate measuring density of the scanning grid and the illumination parameters (including
devices, there are coordinate measuring machines (optical and contact) the measurement of the same element under different illuminations),
[28], articulated arms [29], optical scanners [30], and computer which is particularly important in the context of the reflectivity of the
tomographs [31]. Each of the systems mentioned above and the tech­ measured surface. For technical CT scanners, when selecting measure­
niques they represent present opportunities for artificial intelligence. ment conditions, the system will suggest those resulting from the nature
The support it can provide is vital for the less experienced operator, who of X-rays and the resolution resulting from table rotation, i.e., the
may have doubts about correctly selecting measurement conditions. It number of images per rotation. Once measurement activities are over, AI
will also be more critical for new measured objects for which mea­ can also be very useful in selecting a method for matching the measured
surement programs written by people with expertise do not yet exist. In feature to the data obtained from the CAD file, taking into account that
this regard, it will be essential to support complex measurement tasks this matching can significantly impact the measurement uncertainty of
that include, for example, measurement deviations of form, orientation, the analyzed features. Further calculations can take into account the
location, and run-out deviations. Thus, the tasks of artificial intelligence ambient conditions prevailing during measurement execution, including
will also include selecting measurement conditions and strategies and the difference between the actual temperature and that required ac­
filtering techniques, a common feature of all techniques belonging to cording to the guidelines derived from the standards, as well as the
coordinate metrology. However, each of them still has specific elements, recommended filtering method to match the requirements intended by
which can also be used to determine, for example, techniques related to the designer. The general scheme of this operation is shown in Fig. 3.
machine learning and related numerical procedures. If we are talking Another crucial factor is that the system can expand and supplement
about measurements on CMMs and measuring arms (devices based on the pre-installed database associated with the applications with any new
the collection of points relating to the contact between the measuring tip items and information it has obtained in ongoing operations.
and the surface to be measured), the first element is the loading of CAD Also, at the nano- and microscales, measurements use point col­
data. Using additional knowledge accumulated in the feature database, lecting from the measured surface based on coordinate measurement
appropriate stylus and measuring tips are selected. It is also proposed to technology. The main difference from the macro scale is the approach to
position the feature in the measurement space of a specific device, using the measurement result. In this case, the point cloud is not used to
its data and information resulting from the last calibration. Another evaluate geometric features but to calculate parameters and generate 2D
element is the measurement strategy. First, the measurement system can and 3D features. Filtering techniques, while based on analogous fun­
suggest the order of measured features so that the overall trajectory of damentals in a mathematical sense, are also aimed at slightly different
the measurement head is optimal from the point of view of time and aspects. They aim to expose the most relevant elements of a given sur­
accuracy parameters. It can also propose which measurement mode face and the parameters that prove its functionality. They are indepen­
should be selected (scanning, point-to-point) along with scanning speeds dent of the measuring instruments used, many of which have been
and trajectory. Finally, it can give the minimum and recommended developed in recent years [32]. Although these techniques use different
number of measurement points (for point mode), which will compro­ physical principles (as is evident when considering optical techniques to
mise measurement time and reliable results. The procedures for coor­ analyze surface roughness), two primary groups of methods are distin­
dinate optical scanners and CT scanners are similar. Differences are due guished here: contact (contact profilometry [33] and non-contact, based

Fig. 3. An idea of AI supporting measurements in macro scale.

5
M. Wieczorowski et al. Measurement 217 (2023) 113051

on electromagnetic waves. The decision of which of these two groups metrology, filtering focuses on functional effects. There are many ways
can be subject to artificial intelligence support can be made, mainly to perform this task, including Gaussian filters (linear and robust) [44],
since the surfaces and results obtained can differ significantly [34]. morphological filters [45,46], as well as those using splines [47] or
Optical techniques themselves (such as focus variation microscopy [35], wavelets [48]. The multiscale analysis is also gaining popularity [49].
coherence scanning interferometry [36], phase-shifting interferometric The choice of an appropriate method for discriminating the components
microscopy [37], confocal microscopy [38], point autofocus probe [39], of inequalities is an excellent example of using artificial intelligence.
or confocal chromatic probe [40]) also require special attention, since Especially considering the multitude of possibilities, support from this
even just a change in illumination can produce a completely different site for any operator is invaluable. After filtering, parameters are
image from a surface and thus different measurement results. Increas­ calculated, and functions are generated to describe the surface at this
ingly, the idea of measuring surface irregularities using computed to­ scale. The assistance in this aspect is related to selecting functionally
mography [41,42], traditionally used on the macro scale, is gaining essential parameters [50]. A diagram of the entire process is shown in
ground. Thus, in the case of measurements related to surface irregu­ Fig. 4.
larities, the measurement process assisted by artificial intelligence be­ Based on the parameters and functions defined above, it is possible to
gins with the input of data about the surface, regarding how it is made, verify the functionality of an object at all scales, also conducted in an
its functionality, material, etc. Having these data, the measurement automated manner, in which the operator receives the final decision in a
system can either propose a given measurement technique particularly zero-one system. An additional advantage of conducting a measurement
suited to the application or exclude another due to its adverse effect on process supported by artificial intelligence is the possibility of using
the reliability of the data obtained (e.g., contact profilometry mea­ external data, such as the instrument and the environment, in line with
surements for nominally soft material). Once the measurement tech­ the idea of information-rich metrology [51] and feature extraction [52].
nique is selected, the measurement conditions related to tips or lenses There is another very important aspect of using artificial intelligence
and illumination (mode of incidence, intensity, polarization, etc.) are in metrology. AI has the potential to revolutionize the field of metrology
prepared. If the measurements require illuminations of different in­ by developing new structures of measuring instruments and systems e.g.
tensities, then the system can suggest this to the operator. It is also by utilizing techniques such as machine learning and computer vision.
necessary to specify the measurement field; here, the system can propose Machine learning algorithms can be used to analyze large amounts of
the field and the nesting index based on the expected parameter values. data and identify patterns that can be used to improve the performance
Most optical measurements use stitching [43], in which case the artifi­ of measuring instruments and systems. For example, AI can be used to
cial intelligence-supported system can suggest its parameters or even develop models that can predict the behavior of a system under different
realize the measurement based on flexible stitching, in which the se­ conditions, allowing for more accurate measurements. As it was
lection of the measurement field is carried out in real-time based on the described by Sapozhnikova et al. [53] it is crucial to introduce redun­
identification of surface characteristics defined before the measurement dancy into measuring system under development. It can be of a space
or detected during the measurement. This also has the advantage that if (structural), time / frequency, functional (informational), or combined
the desired components are identified at a specific measurement field, character. AI can be applied at the stage of both development and
and the measurement is performed with stitching from the surface, it can operation. This forms and idea of “metrological self-check”. In tradi­
be automatically completed earlier without performing all the individ­ tional metrology, measurements are often validated by comparing them
ual scans. to a standard, which can be time-consuming and costly. AI-based sys­
Once the data are obtained in microscale and nanoscale coordinate tems can be trained to perform self-validation by analyzing their own

Fig. 4. An idea of AI supporting measurements in micro scale.

6
M. Wieczorowski et al. Measurement 217 (2023) 113051

measurements and comparing them to previous measurements to iden­ external software for multilayer diffraction surfaces. For optical distance
tify any discrepancies. Self-monitoring is another key aspect of measurements, Amin and Riza presented a laser sensor in which they
metrology where AI can be used. AI-based systems can be trained to proposed a supervised machine learning algorithm to improve the ac­
detect and correct errors, which can further improve their accuracy and curacy of the operating range [56]. Yin et al. [57] proposed using su­
reliability. Additionally, AI-based systems can be programmed to self- pervised machine learning, called deep learning, for the phase-tracking
monitor and adjust as needed, which can help to reduce the need for process in fringe projection-based measurement systems. Hoffmann and
human intervention and improve the efficiency of measurement sys­ Elster [58] used deep learning with neural networks to solve the inverse
tems. This can lead to more efficient and effective measurement systems problem in tilted-wave interferometry (TWI). Wu and Fang [59]
that require less human intervention and are able to self-monitor and collected proposals for signal processing presented in various papers,
adjust as needed. AI can be also used to develop measurement systems highlighting advantages and disadvantages and potential directions for
that can perform self-calibration. These systems can analyze their own the further evolution of the methods. Among these methods, they
measurements and make adjustments as needed to ensure that they are pointed to AI-based methods as new ones gaining popularity, pointing
operating within the desired range. This can help to ensure that mea­ out the strong correlation between the accuracy of machine learning
surements are accurate and reliable, and it can also help to reduce the algorithms and the accuracy of initial network learning. A promising
need for manual calibration. technique for the quality control of surfaces fabricated by additive
Metrology 4.0 includes also the use of multisensor solutions, with the manufacturing using scattered light and machine learning was pre­
simpliest version as an optical-contact CMM, which combines a contact sented by Liu et al. [60]. They proposed a simple optical system based on
head with various tips and microscope optics. This allows for mea­ an unsupervised machine learning model as an encoder. Such an optical
surements to be made using both contact and optics. As technology system with the presented algorithm can serve as a fast quality control
advances, these devices are becoming more compact, with the ability to system for individual layers produced during 3D printing. In another
measure surface irregularities at smaller scales and even with interfer­ paper, Liu and Senin [61] presented a prototype of a measurement
ometric optical heads. The effectiveness of using multiple sensors, system to verify a cascaded machine learning model when reconstruct­
especially at different scales, and whether it’s more efficient to use ing surface topography. Artificial intelligence methods are also valuable
separate devices, is still under consideration, as it depends on the spe­ for holographic measurements [62] and scatterometry [63].
cific measurement task and location. Laboratories may prefer universal In this paper, a novel approaches are proposed. In interferometry,
solutions while production may require specialized ones. Still, also here many interference images are recorded, subject to subsequent decoding
AI can be very useful. It can be used to improve the performance of to determine the measured geometric quantities. Interferometers are the
multidimensional measurements, which involve measuring multiple basis for modern interference and holographic microscopes. Together
characteristics or variables at the same time. This type of measurement with gas or semiconductor lasers, they provide submicron spatial mea­
is particularly useful in fields such as manufacturing, where multiple surement resolution. The most often practical applications in metrology
properties of a product or process need to be monitored and controlled. have found Michelson and Mach-Zehnder type interferometers (see
They can perform real-time monitoring and control of multiple char­ Fig. 5). The fringe’s interference signal or image is a common feature of
acteristics or variables. These systems can use machine learning algo­ interferometer-based measurement devices. Data analysis requires
rithms to analyze the data from multiple sensors and make adjustments phase detection and tracking, in which information about the measured
in real-time to optimize the performance of a process or product. AI- geometric quantities is encoded. Most interferometer-based microscopes
based multidimensional measurement systems can also be integrated record data in the form of images. The decoding of the information re­
with other technologies such as Internet of Things (IoT) and cloud quires its analysis and processing. Due to the number of images
computing to provide remote monitoring and control capabilities. This collected, analyzing individual measurement steps is time-consuming
can be particularly useful in industries where measurements need to be and requires an extensive processing algorithm. An example of an
made in remote locations or in harsh environments. A vital part of this is image of fringes acquired in the PSI (Phase-Shifting Interferometric)
also data fusion, that enables the characterisation of an object using measurement system, based on self-built Michelson-type interferometer
multiple datasets collected by various sensors [54]. (see Fig. 6.), is shown in Fig. 7. This is an example image formed after the
reflection of light from a measured object (ceramic roundess standard,
4. The use of artificial intelligence methods in the analysis of ball ∅30 mm).
interference fringes The phase required to determine and track the reciprocal optical
distance of the measured object to the reference mirror corresponds to
In addition to the examples and options above, artificial intelligence the zero-order fringe in the center of the image. The granularity
can assist in signal analysis to make the measurement more efficient and (speckle) visible in the image makes it impossible to use point detectors -
reliable. This chapter presents a novel proposal for using machine image analysis is necessary. The most common and simplest is to
learning in interferometry. This paper describes the data processing analyze, straightforwardly, each recorded image and combine the phase.
methods and their assumptions while experiments are still in progress. In phase-shifting interferometry (PSI), tracking the phase with a small
AI application in measuring industrial parts needs a vast amount of data measurement step becomes crucial, significantly increasing the neces­
collected for the learning process and algorithm preparation. There is a sary number of acquired images and thus increasing processing time.
project in which we work on a database for particular applications for One method of determining this is the “excess fraction” method. This
the future. method takes advantage of the fact that the square of the fringe diameter
Laser measurement systems are capturing large data sets in increases linearly with the order of the fringe:
increasingly shorter times, hence the challenge of analyzing them
D2j = f (j) (1)
quickly and accurately. In this area, artificial intelligence and, in
particular, machine learning (ML) is applicable for the fast analysis of Image processing is required to determine the diameter of the fringes
images acquired with optical scanning. The interaction of light with to use this method. Due to typical optical interference, the image is
matter is a complex phenomenon. It leads to mechanisms of absorption, subjected to filtering steps. Fig. 8 shows an example of the pixel intensity
reflection, scattering, etc. It also depends on many external factors. The distribution in the center of the image, where the phase is located.
application of machine learning algorithms in the field of optical In Fig. 6, the gray line indicates the raw distribution, the dashed line
metrology requires a complex network learning process. Significant is the distribution after filtering using an averaging filter with a 15-pixel
contributions to the field of ML applications in optics were presented by window, and the black line is the distribution after filtering the dashed
Kulce et al. [55]. They described using of networks learned through

7
M. Wieczorowski et al. Measurement 217 (2023) 113051

Fig. 5. Schemes of the interferometers: a) Michelson, b) Macha-Zehnder: λ – coherent light sources, BS – beamsplitters, L – lenses, M – mirrors, CCD – CCD cameras.

Fig. 7. Example of an interference pattern (480x640 pixels) obtained by a

measurement system based on a Michelson-type interferometer from Fig. 6.

phase values to the images of fringes determined using the “excess

Fig. 6. Optical layout of the self-built Michelson-type interferometer: λ – diode fraction” method. In this way, for other applications, the machine
laser with wavelength 532 nm, P- linear polarizer, L1,2,4,5 - achromatic lenses,
learning algorithm, without detailed and complex image processing,
L3 - microscope objective, M - reference mirror with flatness λ/10, BS - non-
could indicate the determined phase changes during measurements.
polarizing beamsplitter, λ/8 - liquid phase retarder, O - object to measure,
PBS - polarizing beamsplitter, CCD1, 2 - CCD cameras.
This approach guarantees a reduction in computation time and, oper­
ating in real-time on images from CCD detectors, reducing the recorded
data.
line with a spline function with 50 degrees of freedom. The points
A second approach, equally effective, is to use the phenomenon of
indicate the maxima. Fig. 9 shows the relationship D2j = f(j).
interference in physical terms. Well-known physical formulas describe
Since the phase in the example image is encoded in the zero-order interference images in an optical system. In this approach, a neural
fringe, the phase fragment j for D2j = 0 can be determined from linear network built on supervised machine learning can determine the phase
regression analysis. Since the fringes are numbered starting from j + 1, in measurement images based on comparisons with images previously
the phase fragment sought is e = 1 − OX. For the case at hand, e = generated based on known values of wave phase shifts. This method
1 − 0.77 = 0.23. Multiplying the phase (e) by half the wavelength de­ bypasses the previously outlined need for calibration and prior pro­
termines the optical distance in metric units. For example, for a wave­ cessing of images using the “straightforward” method. It should be noted
length of 532 nm (green laser), we get a distance of 61 nm from the here that the above concepts are different from those proposed in the
above analysis. Acquiring a bunch of images (24 k typically) during literature.
scanning is necessary to track the phase in the PSI method. This process In phase-shifting interferometry (PSI), commonly used in the mea­
can be accelerated using machine learning algorithms. For this purpose, surement of surface irregularities in microscopy or profilometry, among
two approaches are proposed. Using the above-mentioned “straightfor­ others, the determination and tracking of the phase can also be imple­
ward” method applied to an interference measurement of surface mented using deep convolutional neural networks (CNNs). For this
topography of a standard, the neural network can be taught by assigning purpose, two elements are required: a set of conventionally true data (e.

8
M. Wieczorowski et al. Measurement 217 (2023) 113051

Fig. 8. Distribution of pixel intensities along the horizontal line at the height of the center of the image from Fig. 5.

The use of the techniques described in the paper makes it possible to

support the operator in making decisions on many important aspects of
the measurement process. The system can assist in the selection of stylus
or lenses, the determination of measurement conditions, and actively
participate in the processing of the obtained measurement data. It can
assist in assigning the proper filtration technique to a specific applica­
tion and in selecting surface roughness parameters for specific applica­
tions. As shown in the example of interferometry, specific measurement
techniques can improve the accuracy parameters of analyzed images
and reduce the time to obtain measurement results. The coming years
will probably show other application fields for artificial intelligence in
metrology. This is especially important with the ever-increasing so­
phistication of the measuring instruments and software in use, which
has more and more options requiring advanced knowledge and experi­
ence on the operator’s part. Furthermore, since this is becoming
increasingly difficult in industrial practice, the use of artificial intelli­
gence methods in the future as a support tool will be indispensable. It
Fig. 9. Relationship between the square of the diameter of the interference
can be assumed that AI will become an extension of humans themselves,
fringes as a function of the order of the fringe j.
so instead of talking about artificial intelligence, it can be seen as
augmented intelligence [64].
g. labeled images) and a loss function. Images from simulations based on This paper presents the possibilities of applying artificial intelligence
wave interference equations known in physics can be used to determine to the metrology of geometric quantities, which in recent years has been
the set of correct interferograms, or previous phase calculations of based on coordinate measurement technology. Areas where AI will be
authentic images from measurements might be applied. In addition, the able to assist both the designer and technologist as well as the metrol­
images, in either case, can contain noise and parasitic interference and ogist in their daily work are shown. The practical implications of using
thus may serve to evaluate the accuracy of phase prediction using ma­ AI procedures, among which are the ideas described in this paper, will
chine learning. For example, the loss function for images of fringes can be a much faster preparation of measurement systems for operation and
be performed using a modified version of the mean absolute error a better reliability of the measurement results obtained. Operator ease of
(MAE), considering the difference between the phase values from the use is another feature of systems working with artificial intelligence. As
generated images and those predicted by machine learning. Our first an example application developed by the authors, the use of AI pro­
experimental results for 20,000 images used for training and 4,000 for cedures to evaluate the image of interference striations is presented. Its
testing encompassed a pre-trained ResNet convolutional neural future use is in an instrument for the analysis of surface irregularities
network. The state-of-the analytical phase-determination approach and shape errors, which is being developed as part of a separate project.
shows that the neural network approach achieves an MAE of 0.028, What might the future look like? According to Tegmark [65], the
corresponding to the relative RMSE of about 6 %. In contrast, the exact future of AI research is to build a general artificial intelligence that could
method is slightly worse, with an MAE of 0.31, corresponding to the achieve much more advanced goals, including learning. Based on the
relative RMSE of about 30 %. Processing time for 4,000 images may take examples of using AI in games, it can be seen that the most significant
160 s, while CNN only takes 60 s. Further experimental investigations progress was achieved when the system was equipped only with rules,
are in progress. without previous human knowledge based on examples as training data.
Perhaps this will also result in procedures in metrology that are better
5. Conclusions and future trends and more effective than those used today, even if their content is not
fully known to us. Further into the future, there is also the continuous
Artificial intelligence and related procedures are becoming increas­ improvement of AI procedures in coordinate metrology, including ac­
ingly common in the world around us. This includes issues of metrology tions suggested directly by measurement systems on the basis of previ­
of geometric quantities at various scales, from nano and micro to macro. ous experience, both based on databases created globally (by

9
M. Wieczorowski et al. Measurement 217 (2023) 113051

manufacturers and users of measurement equipment) and locally by the [17] G. Genty, L. Salmela, J.M. Dudley, D. Brunner, A. Kokhanovskiy, S. Kobtsev, S.
K. Turitsyn, Machine learning and applications in ultrafast photonics, Nat.
user himself. Here, naturally, the problem of controlling the develop­
Photonics. 15 (2021) 91–101, https://doi.org/10.1038/s41566-020-00716-4.
ment of AI and the procedures it generates may arise as a limitation, [18] M. Liu, C.F. Cheung, N. Senin, S. Wang, R. Su, R. Leach, On-machine surface defect
which is perhaps the most serious recognised threat at present. Never­ detection using light scattering and deep learning, Journal of the Optical Society of
theless, the practical implications of AI are so promising that work on America A 37 (9) (2020) B53, https://doi.org/10.1364/JOSAA.394102.
[19] A. Aamodt E. Plaza E., Case-based reasoning: foundational issues, methodological
them is a necessity of the coming years. variations, and system approaches, AI Commun 7, 1 (1994) 39–59.
[20] E. Olsson, P. Funk, A. Andersson, Case-based reasoning applied to geometric
CRediT authorship contribution statement measurements for decision support in manufacturing, International Journal of
System Assurance Engineering and Management 4 (3) (2013) 223–230, https://
doi.org/10.1007/s13198-013-0160-y.
Michal Wieczorowski: Conceptualization, Data curation, Formal [21] M. Wieczorowski, N. Swojak, P. Pawlus, A. Pereira, The use of drones in modern
analysis, Funding acquisition, Investigation, Methodology, Project length and angle metrology, in: P. Sniatala, S.S. Iyengar, A. Bendarma, M. Klosak
(eds.), Modern Technologies Enabling Safe and Secure UAV Operation in Urban
administration, Writing – original draft, Writing – review & editing. Airspace, IOS Press (2021) 125 - 140, DOI:10.3233/NICSP210013.
Dawid Kucharski: Conceptualization, Data curation, Formal analysis, [22] M. Wieczorowski, A. Cellary, R. Majchrowski, The analysis of credibility and
Methodology, Writing – original draft, Writing – review & editing. reproducibility of surface roughness measurement results, Wear 269 (5–6) (2010)
480–484.
Pawel Sniatala: Funding acquisition, Formal analysis, Writing – orig­ [23] N. Swojak, M. Wieczorowski, M. Jakubowicz, Assessment of selected metrological
inal draft. Pawel Pawlus: Methodology, Writing – original draft, properties of laser triangulation sensors, Measurement 176 (2021), 109190,
Writing – review & editing. Grzegorz Krolczyk: Project administration, https://doi.org/10.1016/j.measurement.2021.109190.
[24] A. Gaska, W. Harmatys, P. Gaska, M. Gruza, K. Gromczak, K. Ostrowska, Virtual
Writing – original draft, Writing – review & editing. Bartosz Gapinski:
CMM-based model for uncertainty estimation of coordinate measurements
Visualization, Writing – original draft, Writing – review & editing. performed in industrial conditions, Measurement 98 (2017) 361–371, https://doi.
org/10.1016/j.measurement.2016.12.027.
Declaration of Competing Interest [25] R. Su, R. Leach, Physics-based virtual coherence scanning interferometer for
surface measurement, Light: Advanced Manufacturing 2, 2 (2021) 120-135,
https://doi.org/10.37188/lam.2021.009.
The authors declare that they have no known competing financial [26] Z. Peng, M. Huang, Y. Zhong, L. Chen, G. Liu, A new method for interoperability
interests or personal relationships that could have appeared to influence and conformance checking of product manufacturing information, Computers &
Electrical Engineering 85 (2020) 106650, https://doi.org/10.1016/j.
the work reported in this paper. compeleceng.2020.106650.
[27] M. Ren, L. Kong, L. Sun, C. Cheung, A curve network sampling strategy for
Data availability measurement of freeform surfaces on coordinate measuring machines, IEEE
Transactions on Instrumentation and Measurement 66 (11) (2017) 3032–3043,
https://doi.org/10.1109/TIM.2017. 2717283.
No data was used for the research described in the article. [28] S. Hammad Mian, A. Al-Ahmari, New developments in coordinate measuring
machines for manufacturing industries, Int. J. Metrol. Qual. Eng. 5 (2014) 101,
https://doi.org/10.1051/ijmqe/2014001.
References [29] K. Ostrowska, A. Gąska, R. Kupiec, J. Sładek, K. Gromczak, Verification of
articulated arm coordinate measuring machines accuracy using lasertracer system
[1] J. Stenger, J.H. Ullrich, Units Based on Constants: The Redefinition of the as standard of length, MAPAN 31 (2016) 241–256, https://doi.org/10.1007/
International System of Units, Annu. Rev. Condens. Matter Phys. 7 (2016) 35–59, s12647-016-0176-2.
https://doi.org/10.1146/annurev-conmatphys-031115-011311. [30] B. Gapinski, M. Wieczorowski, L. Marciniak-Podsadna, N. Swojak, M. Mendak, D.
[2] R.J. Hocken, P.H. Pereira (Eds.), Coordinate Measuring Machines and Systems, Kucharski, M. Szelewski, A. Krawczyk, Use of white light and laser 3d scanners for
CRC Press, Boca Raton, 2012. measurement of mesoscale surface asperities, in: Advances in Manufacturing II,
[3] B. Schwartz, The Paradox of Choice, Harper Collins Publishers, 2016. V.5, Metrology and Measurement Systems, M. Diering, M. Wieczorowski, C.A.
[4] M. Wieczorowski, D. Kucharski, P. Sniatala, G. Krolczyk, P. Pawlus, B. Gapinski, Brown (eds.), Springer (2019) 239-256.
Theoretical considerations on application of artificial intelligence in coordinate [31] B. Gapinski, M. Wieczorowski, M. Grzelka, P.A., Alonso A.B. Tome, The application
metrology, 6th International Conference on Nanotechnology for Instrumentation of micro computed tomography to assess quality of parts manufactured by means
and Measurement, NanofIM 2021, DOI:10.1109/NanofIM54124.2021.9737344. of rapid prototyping, Polymers 62, 1 (2017) 53-59, DOI: 10.14314/polimery.
[5] M. Khanafer, S. Shirmohammadi, Applied AI in instrumentation and measurement: 2017.053.
The deep learning revolution, IEEE Instrumentation & Measurement Magazine 23 [32] T.G. Mathia, P. Pawlus, M. Wieczorowski, Recent trends in surface metrology,
(6) (2020) 10–17, https://doi.org/10.1109/MIM.2020.9200875. Wear 271 (3–4) (2011) 494–508.
[6] C. Alippi, A. Ferrero, V. Piuri, Artificial intelligence for instruments and [33] K. Grochalski, M. Wieczorowski, B. Jakubek, Influence of thermal disturbances on
measurement applications, IEEE Instrum. Meas. Mag. 1 (2) (1998) 9–17. profilometric measurements of surface asperities, Measurement 190 (2022)
[7] A. Börner, H.-W. Hübers, O. Kao, F. Schmidt, et al., Sensor Artificial intelligence 110694, https://doi.org/10.1016/j.measurement.2021.110694.
and its application to space systems - A White Paper, https://www.user.tu-berlin. [34] P. Pawlus, R. Reizer, M. Wieczorowski, Comparison of results of surface texture
de/peter.jung/papers/Boerner-White_Paper_Sensor_AI.pdf. measurement obtained with stylus methods and optical methods, Metrology and
[8] S. Shirmohammadi, A. Ferrero, Camera as the instrument: The rising trend of vision Measurement Systems 25 (3) (2018) 589–602, https://doi.org/10.24425/123894.
based measurement, IEEE Instrum. Meas. Mag. 17 (3) (2014) 41–47. [35] R. Danzl, F. Helmli, S. Scherer, Focus variation—a robust technology for high
[9] M. Khanafer, S. Shirmohammadi, Applied AI in instrumentation and measurement: resolution optical 3D surface metrology, J Mech. Eng. 57 (2018) 245–256.
The deep learning revolution, IEEE Instrumentation & Measurement Magazine 23 [36] R. Windecker, P. Haible, H.J. Tiziani, Fast coherence scanning interferometry for
(6) (Sept. 2020) 10–17, https://doi.org/10.1109/MIM.2020.9200875. measuring smooth, rough and spherical surfaces, Journal of Modern Optics 42
[10] P. Pouladzadeh, S. Shirmohammadi, R. Al-Maghrabi, Measuring calorie and (1995) 2059–2069.
nutrition from food image, IEEE Trans. Instrum. Meas. 63 (8) (2014) 1947–1956. [37] P. de Groot, Principles of interference microscopy for the measurement of surface
[11] J. Zhou, F. Malric, S. Shirmohammadi, A new hand measurement method to topography, Advances in Optics and Photonics 7 (2015) 1–65.
simplify calibration in CyberGlove-based virtual rehabilitation, IEEE Trans. [38] H. Tiziani, M. Wegner, D. Steudle, Confocal principle for macro- and microscopic
Instrum. Meas. 59 (10) (2010) 2496–2504. surface and defect analysis, Opt. Eng. 39 (2000) 32–39.
[12] S. Pouyanfar, S. Sadiq, Y. Yan, H. Tian, Y. Tao, M. Presa Reyes, M.-L. Shyu, S.- [39] H. Fukatsu, K. Yanagi, Development of an optical stylus displacement sensor for
C. Chen, S.S. Iyengar, A survey on deep learning: Algorithms, techniques, and surface profiling instruments, Microsyst. Technol. 11 (2005) 582–589.
applications, ACM Computing Surveys 51 (5) (2018) 92. [40] H.J. Tiziani, H.-M. Uhde, Three-dimensional image sensing by chromatic confocal
[13] M. Coskun, O. Yildirim, A. Ucar, Y. Demir, An overview of popular deep learning microscopy, Appl. Opt. 33 (1994) 1838–1843.
methods, European J. Technic. 7 (2) (2017) 165–176. [41] B. Gapinski, M. Wieczorowski, L. Marciniak-Podsadna, A. Pereira Dominguez,
[14] M.Z. Alom, T.M. Taha, C. Yakopcic, S. Westberg, P. Sidike, M.S. Nasrin, M. Hasan, L. Cepova, A. Martinez Rey, Measurement of surface topography using computed
B.C. Van Essen, A.A.S. Awwal, V.K. Asari, A state-of-the-art survey on deep tomography, in: A. Hamrol, O. Ciszak, S. Legutko, M. Jurczyk (Eds.), Advances in
learning theory and architectures, Electronics 8 (292) (2019) 1–67. Manufacturing, Springer, 2017, pp. 815–824, https://doi.org/10.1007/978-3-319-
[15] J. Fang, A. Swain, R. Unni, Y. Zheng, Decoding optical data with machine learning, 68619-6_79.
Laser Photon. Rev. 15 (2021) 2000422, https://doi.org/10.1002/lpor.202000422. [42] A. Townsend, L. Pagani, P. Scott, L. Blunt, Areal surface texture data extraction
[16] M. Heizmann, A. Braun, M. Hüttel, C. Kluver, E. Marquardt, M. Overdick, from X-ray computed tomography reconstructions of metal additively
M. Ulrich, Artificial intelligence with neural networks in optical measurement and manufactured parts, Precision Engineering 48 (2017) 254–264, https://doi.org/
inspection systems, - Autom. 68 (2020) 477–487, https://doi.org/10.1515/auto- 10.1016/j.precisioneng.2016.12.008.
2020-0006. [43] F. Marinello, P. Bariani, L. De Chiffre, H.N. Hansen, Development and analysis of a
software tool for stitching three-dimensional surface topography data sets,

10
M. Wieczorowski et al. Measurement 217 (2023) 113051

Measurement Science and Technology 18 (5) (2007) 1404–1412, https://doi.org/ [55] O. Kulce, D. Mengu, Y. Rivenson, A. Ozcan, All-optical information-processing
10.1088/0957-0233/18/5/028. capacity of diffractive surfaces, Light Sci. Appl. 10 (2021) 25, https://doi.org/
[44] J. Seewig, Linear and robust Gaussian regression filters, J. Phys. Conf. Ser. 13 (1) 10.1038/s41377-020-00439-9.
(2005) 254–257. [56] M.J. Amin, N.A. Riza, Machine learning enhanced optical distance sensor, Opt.
[45] S. Lou, X. Jiang, P.J. Scott, Application of the morphological alpha shape method Commun. 407 (2018) 262–270, https://doi.org/10.1016/j.optcom.2017.09.028.
to the extraction of topographical features from engineering surfaces, [57] W. Yin, Q. Chen, S. Feng, T. Tao, L. Huang, M. Trusiak, A. Asundi, C. Zuo, Temporal
Measurement 46 (2) (2013) 1002–1008, https://doi.org/10.1016/j. phase unwrapping using deep learning, Sci. Rep. 9 (2019) 20175, https://doi.org/
measurement.2012.09.015. 10.1038/s41598-019-56222-3.
[46] P. Pawlus, R. Reizer, A. Letocha, M. Wieczorowski, Morphological filtration of two- [58] L. Hoffmann, C. Elster, Deep neural networks for computational optical form
process profiles, Bulletin of the Polish Academy of Sciences: Technical Sciences 67 measurements, J. Sensors Sens. Syst. 9 (2020) 301–307, https://doi.org/10.5194/
(1) (2019) 107–113, https://doi.org/10.24425/bpas.2019.127339. jsss-9-301-2020.
[47] W. Zeng, X. Jiang, P.J. Scott, A generalised linear and nonlinear spline filter, Wear [59] D. Wu, F. Fang, Development of surface reconstruction algorithms for optical
271 (2011) 544–547, https://doi.org/10.1016/j.wear.2010.04.010. interferometric measurement, front, Mech. Eng. 16 (2021) 1–31, https://doi.org/
[48] X.Q. Jiang, L. Blunt, K.J. Stout, Development of a lifting wavelet representation for 10.1007/s11465-020-0602-6.
surface characterization, Proc. R. Soc. Lond. A. 456 (2000) 2283–2313. [60] M. Liu, N. Senin, R. Leach, Intelligent quality monitoring for additive
[49] C.A. Brown, H.N. Hansen, X.J. Jiang, F. Blateyron, J. Berglund, N. Senin, manufactured surfaces by machine learning and light scattering, in: P. Lehmann,
T. Bartkowiak, B. Dixon, G. Le Goic, Y. Quinsat, W.J. Stemp, M.K. Thompson, P. W. Osten, A. Albertazzi Goncalves (eds.), Opt. Meas. Syst. Ind. Insp. XII, SPIE
S. Ungar, E.H. Zahouani, Multiscale analyses and characterizations of surface (2021) 2, https://doi.org/10.1117/12.2592554.
topographies, CIRP Annals 67 (2) (2018) 839–862. [61] M. Liu, N. Senin, Cascaded machine learning model for reconstruction of surface
[50] P. Pawlus, R. Reizer, M. Wieczorowski, Functional importance of surface texture topography from light scattering, in: P.J. de Groot, R.K. Leach, P. Picart (Eds.), Opt.
parameters, Materials 14 (18) (2021) 5326. Photonics Adv. Dimens. Metrol., SPIE (2020) 28, https://doi.org/10.1117/
[51] N. Senin, R. Leach, Information-rich surface metrology, Procedia CIRP 75 (2018) 12.2555035.
19–26. [62] D. Kucharski, M. Bartczak, Application of digital holographic microscopy to
[52] R.H. Schmitt, D. Wolfschläger, E. Masliankova, B. Montavon, Metrologically evaluate the dynamics of a single red blood cell influenced by low-power laser
interpretable feature extraction for industrial machine vision using generative deep light, Opt. Laser Technol. 142 (2021) 107262, https://doi.org/10.1016/j.
learning, CIRP Annals 71 (1) (2022) 433–436, https://doi.org/10.1016/j. optlastec.2021.107262.
cirp.2022.03.016. [63] D. Kucharski, H. Zdunek, A low-cost, simple optical setup for a fast scatterometry
[53] K. Sapozhnikova, A. Pronin, R. Taymanov, Increasing measurement surface roughness measurements with nanometric precision, Bull. Polish Acad. Sci.
trustworthiness as a necessary part of technology development, Sensors & Tech. Sci. 68 (2020) 485–490. https://doi.org/10.24425/bpasts.2020.133364.
Transducers 251 (4) (2021) 61–69. [64] M. Walsh, The Algorithmic Leader: How to Be Smart When Machines Are Smarter
[54] Z.M. Zhang, S. Catalucci, A. Thompson, et al., Applications of data fusion in optical Than You, Page Two Books Inc, 2019.
coordinate metrology: A review, International Journal of Advanced Manufacturing [65] M. Tegmark, Life 3.0, Penguin Books, 2018.
Technology 124 (2023) 1341–1356, https://doi.org/10.1007/s00170-022-10576-
7.

11