9 Inferential Modeling and

Soft Sensors
Muhammad Asad Ullah Khalid
School of Mechanical Engineering, Chung-Ang University,
Seoul, South Korea

Shahid Aziz
Jeju National University, South Korea

Uzair Khaleeq uz Zaman

Department of Mechatronics Engineering, College of
Electrical and Mechanical Engineering, National University
of Sciences and Technology, Islamabad, Pakistan

9.1 INTRODUCTION TO SOFT SENSORS

Soft sensors are essential analytical tools for industrial process monitoring and control.
The word “soft” is short for software, which indicates that these sensors are software-
based. Since these soft sensors are developed using software, they are less expensive to
create than expensive hardware-based sensors. They can complement the existing
hardware-based sensors, allowing the implementation of smart monitoring networks to
provide better process control. The added advantage of real-time data analysis without
time delay issues, preliminarily found in hardware-based sensors, can significantly
improve control strategies. Historical data of different process variables, both
dependent and independent, is very beneficial for soft sensor design and modeling.
This chapter is quite different from the previously reviewed topic on soft sensor
models, as we will briefly present the data-related context for soft sensor modeling and
focus more on a case study-based approach for the applications of soft sensors in
bioprocess monitoring and control. In the end, it will be concluded with the key rec­
ommendations of soft sensors integration for smart process monitoring and control.
There are many different approaches for modeling a soft sensor which can be
mainly categorized into mechanistic, multivariate statistical, and artificial intelligence-
based modeling (Ghosh et al., 2020). The data-driven models need accurate data of the
plant or process variables to make accurate measurements and decisions for control,
which means that the choice of data is crucial in modeling the soft sensors. Modeling a
soft sensor might involve typically following steps (Luigi et al., 2007):

1. Data classification and pretreatment

2. Assigning variables
DOI: 10.1201/9781003327523-11 139
140 Handbook of Manufacturing Systems and Design

3. Soft sensor model structuring and identification

4. Validation of the soft sensor model

Data should be correctly categorized into different classes after collection based on
the information required so that feature identification is easy. Data contamination is
a serious concern, affecting the model’s performance, and the soft sensor model’s
reliability is compromised. This suggests that the data pretreatment is important
before passing it to a model structure. The data pretreatment might involve nor­
malization, filtration, etc. The characteristics of process data have been discussed by
Ghosh et al. (2020) include problems like missing data, data outliers, and data
collinearity and how to deal with them. They are also briefly discussed here to
provide a complete understanding of the concepts. The typical process flow for soft
sensor modeling is given in Figure 9.1.
Before reviewing the data characteristics, let’s look at the flow diagram of soft
sensor-integrated process monitoring and control shown in Figure 9.2. The diagram
shows that the soft sensor depends on a hardware sensor’s data to determine the
estimated variables. The process generates process variables, which pass through
this soft sensor module comprising a hardware sensor and an estimator. This esti­
mator is our soft sensor design model, which estimates new parameters by ana­
lyzing the historical process data of the hardware sensor to determine new estimated
variables. These new estimated variables can then be used as process monitoring
parameters as well as control parameters. To keep the consistency of the concepts
being discussed, it is important to mention that the estimator or soft sensor model is

FIGURE 9.1 A block diagram of soft sensor modeling process.

FIGURE 9.2 A flow diagram of a soft sensor integrated process monitoring and control.
Inferential Modeling and Soft Sensors 141

entirely dependent on the single/multiple hardware sensors’ data. We previously

described that this data is available in the raw form and is pretreated due to certain
problems. One major problem is data collinearity, which makes it very time-
consuming to select the appropriate data bank from huge data with poor information
management. If not tackled properly, it could seriously affect the performance of
data-driven, for example, artificial neural network (ANN)-based sensors. The plant
personnel working might help acquire the right data due to their working experience
with that particular process control. Then the problem of missing data can also
emerge, which can happen because of power outages due to maintenance, repairs,
and certain other reasons. So, these missing data need to be accommodated.
Sometimes this missing data is ignored within a chronological stream, but at other
times, it is replaced with a filler data set to complete a chronological data stream.
Data outliers are also a prevalent problem. They are mostly the process variable
data points, which are outside a defined or normal range of interest, and are caused
by instrumental errors, plant disturbances, etc. They can negatively affect the soft
sensor model. They can be categorized into different types and handled accordingly
using various data reconciliation and proximity-based methods (Ghosh et al., 2020).
Virtual instruments (VIs) play an important role in modeling soft sensors. These
instruments are software-based measurement and computational blocks that utilize
the existing hardware instruments and computer systems with programmed user
interfaces. These VIs can help design soft sensors as they provide customization
capabilities. They can be utilized according to application-based requirements to
measure and process data. All these VIs are available in a software facility store as
packages for custom soft sensor modeling. LabVIEW by National Instruments (NI)
is a powerful computer program used in many industrial process monitoring and
control applications. It combines with various data acquisition hardware or other
instruments for real-time variables’ data collection. This raw data is then compu­
tationally evaluated, combined with other data, or processed according to certain
logical and mathematical constraints within a LabVIEW program. This LabVIEW
program provides a convenient front panel interface for user data display and control,
while a block diagram represents a backend software that runs the algorithm to
perform tasks in the front panel according to user design applications. This provides a
great degree of comfort to the designers and users as well. The users are mostly
concerned with these front panel tasks, which are relatable to the actual physical
process as it is designed to do so, whereas all the logical and mathematical operations
are carried out by the implemented algorithm using VIs in the block diagram.
Examples of a typical front panel and block diagram are provided in Figure 9.3.

9.2 SOFT SENSOR INFERENTIAL MODELS

Modern-day manufacturing plants are very different from the old ones. They consist
of old equipment, which was built in the 1920s as well as new equipment up to the
2020s. Smart manufacturing (SM) is utilizing data science to improve efficiency at
all levels of the manufacturing cycle. Since it is not possible to get real-time
feedback from all important processes, there is a need to use soft sensors and
inferential models to realize the feedback. Improvements in sensors embedded in
142 Handbook of Manufacturing Systems and Design

FIGURE 9.3 Images of a front panel and block diagram for LabVIEW-based VIs for
sensor-based measurements in a bioprocess ( Khalid et al., 2022).

the manufacturing processes enable inferential models that simulate manufactur­

ability characteristics that affect quality, cost, and ease of assembly. Inferential
modeling techniques automate the process of generating digital twin models through
direct observation, thereby maximizing the benefits of Industry 4.0 technologies. It
enables the automatic modeling of the processes using computational models and
automatically updates models due to the change in environment. Inferential modeling
relies on the input data from various sensors to estimate the output, which lies at the
center of modeling techniques to provide robust, repeatable, and sustainable results.
Data can be efficiently utilized for modeling when available or refined in a meaningful
manner; hence, data collection and training are critical to infer effective soft sensor
models in the manufacturing plant. There are various techniques for modeling soft
sensors. In this section, we will discuss principal component analysis (PCA), partial
least squares (PLS), and ANN techniques to model input data from soft sensors.

9.2.1 PRINCIPAL COMPONENT ANALYSIS

SM requires smart measurement methods to monitor the advanced process control
parameters. Since it is not always possible to measure them, soft sensors are used to
model these processes. PCA is one of the methods to model the huge amount of data
inferred from the soft sensors, which relies on reducing the dimension of the data
set while increasing the probability of retaining the variance of the parent data. PCA
method is preferable when the manufacturing process is linear; otherwise, it is not
recommended. PCA has been widely used in various soft sensor applications (Shi &
Liu, 2006; Pe et al., 2018; Farsang et al., 2015).
Inferential Modeling and Soft Sensors 143

In order to model sensor data, consider a data matrix M = {(X, Y)p; p = 1,

2, …, n}, where n is the number of sample data points, and X = (x1, x2, x3, … , xl),
Y = (y1, y2, y3, … , ym), where l and m represent the number of input and output
variables of soft sensors.
All input variable data X is selected by PCA to filter the irrelevant input vari­
ables from going into the multiscale decomposition stage. Based on the experi­
mental data and trial error, the selected new variables are scaled down to several
components that best explain the data variation. At this stage, new pairs are formed
in the training process, and then the PCA model reconstructs the trained data in the
multiscale reconstruction stage before it is obtained as a model output Y.

9.2.2 PARTIAL LEAST SQUARES

Partial least squares (PLS) is another simple method for inferential modeling of soft
sensor data (Hasnen et al., 2019; Wang et al., 2015). PLS is used to evaluate the
performance of the variable selection methods. Just like PCA, PLS is also limited
only to linear processes. However, some researchers have modified PLS algorithms
to account for the non-linearities subjected to time-varying characteristics experi­
enced by most manufacturing processes in industries (Fu et al., 2017). Since PLS is
based on the variables’ scores instead of the original values of variables, it is
considered more stable than those models that use the original variables.
The mathematical model of the PLS has been explained in detail by some re­
searchers (Fu et al., 2017). The basic idea is to start modeling by centralizing and
normalizing the input and output data matrices. Then the PLS algorithm is used to
decompose the input and output matrices into external models having a component
matrix that consists of score matrices. The detailed breakdown of the PLS matrices
can be found in Fu et al. (2017), Chen et al. (2009), and Poerio and Brown (2020).

9.2.3 ARTIFICIAL NEURAL NETWORKS

ANNs are algorithms that help deep learning to solve advanced data-driven prob­
lems. Deep learning makes use of ANN that behave similarly to the neural networks
in a human brain. A neural network functions when some input data is fed to it. This
data is then processed via layers of perceptrons to produce the desired output.
During the training process, the ANN-based model can learn the complex dynamics
of the manufacturing process through variable examples readily available through
previous experimental data. Unlike PCA and PLS models, the ANN model is well
known to deal with complex nonlinearities, which play an essential role in SM
processes. ANN is used extensively in SM for Industry 4.0 to predict important
process parameters critical for mechanical components. In a recent work (Kazi
et al., 2021), ANN has been used to predict the required cotton fiber filler content
to increase the mechanical properties in a cotton fiber/thermoplastic composite,
i.e., polypropylene. The polyethylene terephthalate production process has been
monitored via soft sensors developed through a feed-forward ANN model
(Gonzaga et al., 2009). ANN-based soft sensors have been utilized to estimate the
apparent viscosity of water-based drilling fluids from the available data about the
144 Handbook of Manufacturing Systems and Design

concentration of additives and temperature of that fluid (Bispo et al., 2017). The
modeling of ANN-based soft sensors can be found in many research works (“Soft
Sensor Model Based on Improved Elman Neural.Pdf”; Ghosh et al., 2020).

9.2.4 SUPPORT VECTOR MACHINES

ANN-based modeling has been successfully implemented in soft sensor modeling,
but they lack the ability for generalization. Therefore, support vector machines
(SVM)-based soft sensor modeling is needed. The basis of SVM is the statistical
learning theory. SVMs are more robust to raw data and have a better ability for
generalization because of the natural enforcement of regularization. To predict key
production indicators in the complex grinding process, a modification of SVM, i.e.,
least squares SVM, has been used as a soft sensor model with strong generalization
ability (Xie et al., 2020). SVM has also been used to develop soft sensor models
for continually assessing quality indicators in the refinery isomerization process of
2,3-dimethylbutane and 2-methyl-pentane mole percentage (Herceg et al., 2021).
The advantages of SVM over ANN are the convergence of SVM-based soft sensors
to the global optimal more efficiently and its ability to be interpreted based on the
training data, which cannot be done in the case of ANN-based soft sensors.

9.3 SOFT SENSORS IN MULTISTAGE SM

SM processes have become more complex and multistage due to several quality
control parameters and customer requirements. In most multistage manufacturing
processes, the process monitoring techniques provide prediction models at the end of
the manufacturing process, which is less practical and costly. Soft sensor models
discussed in the previous section can be useful in challenging modeling problems like
multistage SM processes. This requires continuous online monitoring of processes,
which involves big data handling to predict anomalies in manufacturing processes
early in the multistage SM system. Several soft sensor modeling approaches, such as
ANN surrogate (Bambach et al., 2021) and hybrid sequence-to-sequence recurrent
neural network-deep neural network (Seq2Seq RNN–DNN) (Hong et al., 2022) can
be adapted to model multistage manufacturing processes in which each stage can be
monitored through a different soft sensing approach along with embedded sensors
throughout the processes. ANN surrogate model-based soft sensing approach has
been demonstrated by Bambach et al. (2021) to model the multistage hot forming
process. A hybrid model (Hong et al., 2022) combining several soft sensing models
can be integrated to enhance quality monitoring in multistage SM systems.

9.4 HARD AND SOFT SENSORS FOR SMART BIOREACTORS

This is an important and new topic that distinguishes the contents of this book
chapter from others in similar contexts. Bioreactors and their different types will be
briefly discussed here, most importantly focusing on the hardware and software
sensors used for bioprocess monitoring (Biechele et al., 2015; Busse et al., 2017).
Inferential Modeling and Soft Sensors 145

Then this topic will briefly discuss some soft sensors implemented-bioreactors with
a case study.
Bioreactors are engineered spaces to allow safer and controlled bioreactions. To
briefly describe the main purpose of bioreactor designing is to realize perfect
conditions for these bioreactions, which may involve cell culture and micro­
organism growth for various applications like the production of biofuels, metabolic
products, biowaste management etc. There are many types of bioreactors depending
on the application, to name a few, they are stirred-tank reactor, bubble reactor,
airlift reactor, loop reactor, reactor with immobilized cells, fluidized reactor with
recycling of cells, solid-phase tray reactor, rotary drum bioreactor as discussed by
Carl (2016). They have discussed the advantages and shortcomings of these bior­
eactors’ designs. In this text, only specific sensor-integrated bioreactors will be
discussed. When we talk about key parameters of bioprocess monitoring, they
mostly include some physical and/or chemical environmental parameters, e.g.,
temperature, CO2, and culture media pH. However, biosensors also play a very
important role in determining the state and output of the bioreactor systems. These
biosensors can detect metabolic products, proteins, DNA/RNA structures, or the
microorganisms themselves with very high specificity. For example, a review was
conducted on the usefulness of soft sensing systems in online monitoring of up­
stream industrial bioprocess, focusing on the sensing performance and process’
outputs in terms of production economy (Randek & Mandenius, 2018). A basic
overview of the soft sensing requirements and applications in various industrial
bioprocesses, like recombinant protein production and production of baker’s yeast
has been presented. Similarly, different online analytical techniques like near and,
mid-infrared (N/MIR) spectroscopy with high potential for soft sensor development
have been discussed, which can measure several parameters like metabolic prod­
ucts’ concentrations or cell numbers. Other discussed techniques include in-situ
microscopy (ISM), fluorescence microscopy, high-performance liquid chromatog­
raphy (HPLC), and biosensors. Similarly, they reviewed various upstream biopro­
cess monitoring by soft sensors’ examples, followed by suggesting a step-wise soft
sensor development general methodology for this purpose.

9.4.1 CASE STUDY – SOFT SENSOR FOR BIOPROCESS MONITORING

Metabolic heat-based soft sensor has been used in a fed-batch Escherichia coli
cultivation-based recombinant protein production (green fluorescent protein – GFP)
bioprocess for estimating the specific growth rate and biomass concentration
(Paulsson et al., 2014) as shown in Figure 9.4a. The sequential filtering further
improved the methods like moving average filtering methods, such as low-pass
filters, Savitzky-Golay filters, and extended Kalman filters. They successfully
controlled the feed rate based on the soft sensor’s output estimates. After sequential
filtering, the final model equation obtained was

qfilter 2, n = {(qfilter 1 [n 180 + 1: n]) + min (qfilter 1 [n 180 + 1: n])}/2 …

(9.1)
146 Handbook of Manufacturing Systems and Design

FIGURE 9.4 (a) Metabolic heat soft sensor for bioprocess control, and (b) Estimated
metabolic heat, specific growth rate (µmetabol) and biomass concentration (Xmetabol,n) with
passing cultivation time in pre-set feed rate ( Paulsson et al., 2014).

Similarly, further mathematical modeling was done to derive specific growth rate
µmetabol and biomass concentration Xmetabol,n and the feed rate was controlled using a
conventional proportional-integral (PI) controller. Further details can be found in
the original work (Paulsson et al., 2014). Figure 9.4b shows the estimated outputs of
the soft sensing system for uncontrolled fed-batch cultivation. In the end, the soft
sensor’s performance with hardware-based sensor probes and capacitance on-line
sensor, measuring the same parameters, were in high agreement.
Apart from this, many other soft sensors have been developed for bioprocess
monitoring and control applications. A few of them have been presented in
Table 9.1 for a brief review.
Inferential Modeling and Soft Sensors 147

TABLE 9.1
Soft Sensors Developed for Various Bioprocesses’ Monitoring and Control
Soft Sensors Purpose & Parameters Estimated Reference
a. Biomass concentration from an Recombination protein (GFP) Warth et al.,
online NIR probe, production in E. Coli cultivation, 2010
b. Biomass concentration from titrant mainly for
addition, a. Biomass concentration estimation
c. Specific growth rate from titrant b. Estimation of specific growth rates
addition,
d. Specific growth rate from the NIR
probe, and
e. Specific substrate uptake rate and by-
product rate from online HPLC and
NIR probe signals

Feed-forward ANN, support vector, and Biomass and product concentrations Simutis et al.,
relevance vector regression soft sensor 2013
mode based on glucose/lactose feed
rate and oxygen uptake
Biomass soft sensor in Eppendorf Biomass production and biomass Kager et al., 2017
DASbox® Mini Bioreactor System specific nutrient uptake estimation
using the DASware® control software (e.g., substrate uptake rate)
using mass balancing
Multiphase Artificial Neural Network For biomass concentration of Murugan and
(MANN) based dynamic soft sensor Trichoderma estimated from online Natarajan, 2019
having Nonlinear Auto Regressive sensors data like pH, substrate
with eXogenous input (NARX) models concentration and agitation speed.
to capture the complete dynamics of
lag, log, and stationary phases of the
microbe
The end of the growth phase in a Dynamic control of compressor based Alarcon and
thraustochytrid Industry 4.0 (I4.0) on dissolved oxygen (DO) levels Shene, 2021
fermentation process

9.4.2 CASE STUDY – ADVANCED SENSING STRATEGIES WITH SOFT SENSING

POTENTIAL FOR BIOCHIP SYSTEMS
The authors have worked on state-of-the-art microfluidic organs-on-chips, which are
engineered microchips – sometimes also referred to as biochips, containing living
human organ microtissues in a dynamic culture environment to recapitulate the in-
vivo-like organ functionalities in-vitro (Wu et al., 2020; Leung et al., 2022). They are
recently becoming popular in different fields of study, e.g., toxicology, drug devel­
opment (Caplin et al., 2015; Ching et al., 2021), disease modeling (Wang et al., 2017;
Farooqi et al., 2021; Asif et al., 2021), and personalized medicine (Ingber, 2022).
While they have the potential to replace animal testing models in the traditional drug
148 Handbook of Manufacturing Systems and Design

development pipeline, they still need to improve on continuous processing monitoring

and control strategies due to a lack of technological advancements. The authors have
developed an embedded-sensors gut-bilayer microchip for real-time assessment of
cellular dissolved oxygen (DO) and reactive oxygen species concentrations to analyze
developmental and induced hypoxia (Khalid et al., 2022). The expanded and cross-
sectional views of this biochip are shown in Figures 9.5a and b. Similarly, the bio­
process schematic demonstration and real-time data of the sensors are given in
Figure 9.5c and d, respectively. While this biochip system generates very useful
data of sensors for process monitoring, for a comprehensive analysis, the soft sensors
to estimate metabolic products like glucose uptake and lactate release can also
be implemented using the existing hardware sensors data as inputs to the model in
future studies.
Another study on the microfluidic biochip was conducted by the authors to
analyze the effect of chemotherapeutic drugs (docetaxel and doxorubicin) on a lung
cancer microtissue using the integrated hardware sensors (optical pH sensor,

FIGURE 9.5 A gut bilayer organ on chip for hypoxia bioprocess monitoring: (a) Sensors
(DO, reactive oxygen species, and TEEI) integrated microfluidic biochip expanded view,
(b) Schematic cross-section indicating different components like printed electrochemical
(DO and reactive oxygen species) sensors, TEEI sensor, cell culture area and cell types etc.,
(c) Hypoxia bioprocess schematic, (d) Response of the sensors in real-time for both normoxia
and hypoxia bioprocesses ( Khalid et al., 2022).
Inferential Modeling and Soft Sensors 149

FIGURE 9.6 A lung cancer on chip with integrated sensors for process monitoring and
chemotherapeutic assessment: (a) Schematic representation of the sensor (optical pH sensor,
TEER sensor, and a portable microscope) integrated process, (b) Results of the sensors in
normal process monitoring and response analysis to chemotherapeutic drug Doxorubicin
( Khalid et al., 2020).

Transepithelial electrical resistance [TEER], and portable microscope) approach

(Khalid et al., 2020). The sensor data were collected for normal growth monitoring
and after treatment with different drug concentrations in real-time. A process
schematic is shown in Figure 9.6a, whereas the sensor output response is shown in
Figure 9.6b.
This system can produce enormous data banks of the process parameters using
integrated online monitoring sensors, which could again be used for estimating
relevant secondary parameters. These secondary parameters could be metabolic
products, as previously described, as well as some pathological protein biomarkers
or even DNA or RNA. The visual monitoring capability can allow for geometrical
growth predictions of computer vision-based soft sensor models. The use of soft
sensor models can be extremely beneficial for the complex multi-organ on-chip
systems, where hardware usage is plentiful. The complex multi-organ on-chip
systems provide more data, and data-driven soft sensors can benefit from this for
state estimation and process control.

9.4.3 CONCEPTUAL FRAMEWORK FOR FUTURE SMART BIOPROCESS MONITORING

SYSTEMS
A Hubka-Eder map (Hubka & Eder, 1988) can help build smart, complex bio-
reactor systems, as shown in Figure 9.7, by identifying the contributing individual
 150

FIGURE 9.7 The Hubka-Eder map for smart bioreactor on chip systems.
Handbook of Manufacturing Systems and Design
Inferential Modeling and Soft Sensors 151

functional units like biological components, technical systems, Information man­

agement systems, and human resources. These subunits contribute towards the
transformation of primary inputs into results and estimated parameters. The trans­
formation process can be divided into various execution phases, which convert the
inputs to outputs using hardware and soft sensors.

9.5 CONCLUSION
Data-driven soft sensor models expand the capabilities of existing hardware sensors
and systems in traditional integrated process monitoring and control applications.
They can estimate the parameters which are not compatible with hardware mea­
surement strategies. However, data selection is the most crucial step in modeling soft
sensors. This data should be high-quality and free of noise, disturbances, and mis­
information, for which different data treatment strategies are applied. A few infer­
ential models have been briefly reviewed in this chapter. And some recent literature
on soft sensors in smart multi-stage manufacturing systems and smart bioreactor
systems has also been discussed, which suggest the more widespread use of soft
sensor models in complex bioprocess monitoring and control with a case study.
Similarly, emerging state-of-the-art bioreactor on-chip systems, which use online
monitoring strategies, were explored with the potential for soft sensor integration to
enhance different biotechnological aspects of these systems. Some conceptual soft
sensor integration frameworks have also been presented in this regard.

9.5.1 FUTURE TRENDS (BIG DATA AND DEEP LEARNING REVOLUTION)

Since the data volumes are increasing with the complexity and abundance of the
available hardware, there is a great need to utilize deep learning algorithms. However,
the general rules of data selection apply here as well. But due to enormous instru­
mentation and sensing devices in modern-day plants, the data is generated in big
volumes and variety. This big data needs large storage and a large number of data
processing units to extract useful data features or parametric information. This could
also lead to longer data processing times. Thankfully, the current data processing
systems have high computational capabilities and can process big data in a reasonable
time frame. The second important aspect is to use intelligent deep learning algorithms
to model the soft sensors. A few example models have been discussed in the refer­
ences (Ma et al., 2020; Wang et al., 2020) for the reader’s choice.

REFERENCES
Alarcon, C., Shene, C. 2021. Fermentation 4.0, a Case Study on Computer Vision, Soft
Sensor, Connectivity, and Control Applied to the Fermentation of a Thraustochytrid.
Computers in Industry, 128: 103431.
Asif, A., Park, S. H., Soomro, A. M., Khalid, M. A. U., Salih, A. B. C., Kang, B., Ahmed, F.,
Kim, K. H., Choi, K. H. 2021. Microphysiological System with Continuous Analysis
of Albumin for Hepatotoxicity Modeling and Drug Screening. Journal of Industrial
and Engineering Chemistry, 98: 318–326. doi: 10.1016/j.jiec.2021.03.035
152 Handbook of Manufacturing Systems and Design

Bambach, M., Imran, M., Sizova, I., Buhl, J., Gerster, S., Herty, M. 2021. A Soft Sensor for
Property Control in Multi-Stage Hot Forming Based on a Level Set Formulation of Grain
Size Evolution and Machine Learning. Advances in Industrial and Manufacturing
Engineering, 2: 100041.
Biechele, P., Busse, C., Solle, D., Scheper, T., Reardon, K. 2015. Sensor Systems for
Bioprocess Monitoring. Engineering in Life Sciences, 15(5): 469–488.
Bispo, V. D. da Silva, Scheid, C. M., Calçada, L. A., Meleiro, L. A. C. 2017. Development of
an ANN-Based Soft-Sensor to Estimate the Apparent Viscosity of Water-Based
Drilling Fluids. Journal of Petroleum Science and Engineering, 150: 69–73.
Busse, C., Biechele, P., Vries, I., Reardon, K. F., Solle, D., Scheper, T. 2017. Sensors for
Disposable Bioreactors. Engineering in Life Sciences, 17(8): 940–952.
Caplin, J. D., Granados, N. G., James, M. R., Montazami, R., Hashemi, N. 2015. Microfluidic
Organ-on-a-Chip Technology for Advancement of Drug Development and Toxicology.
Advanced Healthcare Materials, 4(10): 1426–1450.
Carl, F. M. 2016. Bioreactors. Edited by Carl-Fredrik Mandenius. Weinheim, Germany:
Wiley-VCH Verlag GmbH & Co. KGaA.
Chen, K., He, M., Zhang, D. 2009. Sliding-Window Recursive PLS Based Soft Sensing
Model and Its Application to the Quality Control of Rubber Mixing Process.
Communications in Computer and Information Science, 51: 16–24.
Ching, T., Toh, Y. C., Hashimoto, M., Zhang, Y. S. 2021. Bridging the Academia-to-Industry
Gap: Organ-on-a-Chip Platforms for Safety and Toxicology Assessment. Trends in
Pharmacological Sciences, 42(9): 715–728.
Farooqi, H. M. U., Kang, B., Khalid, M. A. U., Salih, A. R. C., Hyun, K., Park, S. H.,
Huh, D., Choi, K. H. 2021. Real-Time Monitoring of Liver Fibrosis through
Embedded Sensors in a Microphysiological System. Nano Convergence, 8(1).
Springer Singapore.
Farsang, B., Balogh, I., Németh, S., Székvölgyi, Z., Abonyi, J. 2015. PCA Based Data
Reconciliation in Soft Sensor Development – Application for Melt Flow Index
Estimation. In Chemical Engineering Transactions, 43: 1555–1560.
Fu, Y., Yang, W., Xu, O., Zhou, L., Wang, J. 2017. Soft Sensor Modelling by Time Difference,
Recursive Partial Least Squares and Adaptive Model Updating. Measurement Science
and Technology, 28(4).
Ghosh, S., Yang, S., Bequette, B. W. 2020. Inferential Modeling and Soft Sensors. In Smart
Manufacturing: Concepts and Methods, 323–351. Elsevier Inc.
Gonzaga, J. C. B., Meleiro, L. A. C., Kiang, C., Filho, R. M. 2009. ANN-Based Soft-Sensor
for Real-Time Process Monitoring and Control of an Industrial Polymerization
Process. Computers and Chemical Engineering, 33(1): 43–49.
Hasnen, S. H., Zabiri, H., Prakash, K. K., Mat, T. T. 2019. Adaptive PLS Inferential Soft
Sensor for Continuous Online Estimation of NOx Emission in Industrial Water-Tube
Boiler. IOP Conference Series: Materials Science and Engineering, 702(1).
Herceg, S., Andrijić, U., Bolf, N. 2021. Support Vector Machine-Based Soft Sensors in the
Isomerisation Process. Chemical and Biochemical Engineering Quarterly, 34(4):
243–255.
Hong, S., An, N., Cho, H., Lim, J., Han, I. S., Moon, I., Kim, J. 2022. A Dynamic Doft
Sensor Based on Hybrid Neural Networks to Improve Early Off-Spec Detection.
Engineering with Computers, 0123456789. Springer London.
Hubka, V., Eder, W. E. 1988. Theory of Technical Systems. Berlin, Heidelberg: Springer
Berlin Heidelberg.
Ingber, D. E. 2022. Human Organs-on-Chips for Disease Modelling, Drug Development and
Personalized Medicine. Nature Reviews Genetics, 0123456789. Springer US.
Inferential Modeling and Soft Sensors 153

Kager, J., Fricke, J., Herwig, C. 2017. A Generic Biomass Soft Sensor and Its Application in
Bioprocess Development. Eppendorf, 357: 1–8, https://www.eppendorf.com/
product‐media/doc/en/228755/Fermentors‐Bioreactors_Application‐Note_357_
DASware_A‐Generic‐Biomass‐Soft‐Sensor‐Its‐Application‐Bioprocess‐Development.
pdf (accessed on 7 June 2023).
Kazi, M. K., Eljack, F., Mahdi, E. 2021. Data-Driven Modeling to Predict the Load vs.
Displacement Curves of Targeted Composite Materials for Industry 4.0 and Smart
Manufacturing. Composite Structures, 258: 113207.
Khalid, M. A. U., Kim, K. H., Salih, A. B. C., Hyun, K., Park, S. H., Kang, B., Soomro, A.
M. 2022. High Performance Inkjet Printed Embedded Electrochemical Sensors for
Monitoring Hypoxia in a Gut Bilayer Microfluidic Chip. Lab on a Chip, 22(9):
1764–1778.
Khalid, M. A. U., Kim, Y. S., Ali, M., Lee, B. G., Cho, Y.-J., Choi, K. H. 2020. A Lung
Cancer-on-Chip Platform with Integrated Biosensors for Physiological Monitoring and
Toxicity Assessment. Biochemical Engineering Journal, 155: 107469.
Leung, C. M., Haan, P., Bouchard, K. R., Kim, G. A., Ko, J., Rho, H. S., Chen, Z., et al.
2022. A Guide to the Organ-on-a-Chip. Nature Reviews Methods Primers, 2(1).
Springer US.
Luigi, F., Graziani, S., Rizzo, A., Xibilia, M. G. 2007. Soft Sensors for Monitoring and Control
of Industrial Processes. Advances in Industrial Control. London: Springer London.
Ma, Y., Liu, S., Xue, G., Gong, D. 2020. Soft Sensor with Deep Learning for Functional
Region Detection in Urban Environments. Sensors, 20(12): 1–17.
Murugan, C., Natarajan, P. 2019. Estimation of Fungal Biomass Using Multiphase Artificial
Neural Network Based Dynamic Soft Sensor. Journal of Microbiological Methods,
159: 5–11.
Paulsson, D., Gustavsson, R., Mandenius, C. F. 2014. A Soft Sensor for Bioprocess
Control Based on Sequential Filtering of Metabolic Heat Signals. Sensors, 14(10):
17864–17882.
Pe, E. A. G., Oliveira, V. A., Cruvinel, P. E. 2018. Soft-Sensor Approach Based on Principal
Components Analysis to Improve the Quality of the Application of Pesticides in
Agricultural Pest Control. 95–100.
Poerio, D. V., Brown, S. D. 2020. Localized and Adaptive Soft Sensor Based on an Extreme
Learning Machine with Automated Self-Correction Strategies. Journal of Chemometrics,
34(7): 0–2.
Randek, J., Mandenius, C. F. 2018. On-Line Soft Sensing in Upstream Bioprocessing.
Critical Reviews in Biotechnology, 38(1): 106–121.
Shi, J., Liu, X.-G. 2006. Product Quality Prediction by a Neural Soft-Sensor Based on MSA
and PCA. International Journal of Automation and Computing, 3(1): 17–22.
Simutis, R., Galvanauskas, V., Levisauskas, D., Repsyte, J., Vaitkus, V. 2013. Comparative
Study of Intelligent Soft-Sensors for Bioprocess State Estimation. Journal of Life
Sciences and Technologies, 1(3): 163–167.
Wang, B., Wang, Z., Chen, T., Zhao, X. 2020. Development of Novel Bioreactor Control
Systems Based on Smart Sensors and Actuators. Frontiers in Bioengineering and
Biotechnology, 8: 7.
Wang, L., Tao, T., Su, W., Yu, H., Yu, Y., Qin, J. 2017. A Disease Model of Diabetic
Nephropathy in a Glomerulus-on-a-Chip Microdevice. Lab on a Chip, 17(10):
1749–1760.
Wang, Z. X., He, Q. P., Wang, J. 2015. Comparison of Variable Selection Methods for PLS-
Based Soft Sensor Modeling. Journal of Process Control, 26: 56–72.
154 Handbook of Manufacturing Systems and Design

Warth, B., Rajkai, G., Mandenius, C.-F. 2010. Evaluation of Software Sensors for On-line
Estimation of Culture Conditions in an Escherichia Coli Cultivation Expressing a
Recombinant Protein. Journal of Biotechnology, 147(1): 37–45.
Wu, Q., Liu, J., Wang, X., Feng, L., Wu, J., Zhu, X., Wen, W., Gong, X. 2020. Organ-on-a-
Chip: Recent Breakthroughs and Future Prospects. BioMedical Engineering Online,
19(1): 1–19.
Xie, W., Wang, J. S., Xing, C., Guo, S. S., Guo, M. W., Zhu, L. F. 2020. Adaptive Hybrid
Soft-Sensor Model of Grinding Process Based on Regularized Extreme Learning
Machine and Least Squares Support Vector Machine Optimized by Golden Sine Harris
Hawk Optimization Algorithm. Complexity, 2020.