Computers and Chemical Engineering 136 (2020) 106756

Contents lists available at ScienceDirect

Computers and Chemical Engineering

journal homepage: www.elsevier.com/locate/compchemeng

Data-driven process monitoring and fault analysis of reformer units in

hydrogen plants: Industrial application and perspectives
Ankur Kumar∗, Apratim Bhattacharya, Jesus Flores-Cerrillo
Linde Digital, Linde Technology Center, 175 East Park Dr, Tonawanda, NY 14150, United States

a r t i c l e i n f o a b s t r a c t

Article history: Reformer boxes are complex, integrated, and high-temperature units, subject to various failures during
Received 9 September 2019 continuous operations for extended time periods. Challenges in the development of high-fidelity first
Revised 30 December 2019
principle models, despite easy availability of process measurements motivated the development of data-
Accepted 23 January 2020
driven, automated fault detection (FD) systems. Paucity of plant-wide implementation of FD technologies
Available online 28 January 2020
in the chemical industry, accentuates the absence of relevant practical guidelines and best practices. In
Keywords: this paper, a trivially replicable FD system has been developed for large-scale industrial reformer boxes
Steam-methane reformers of hydrogen manufacturing units. Actual process data from plant historian has been used for training and
Process monitoring validation of a novel model, developed using a combination of partial least squares regression and prin-
Abnormality detection cipal components analysis. Abnormalities based on several important measurements around the reformer
Data-driven modeling were identified. Explicit algorithmic details and insights obtained during development of the expert sys-
tem have been provided for ease of replication and adaptability.
© 2020 Elsevier Ltd. All rights reserved.

1. Introduction
Large scale industrial processes warrant increasing demands on
process safety, uniformity in production quality, and overall plant
reliability. To achieve these objectives, real-time process monitor-
ing is employed for early detection of abnormalities in processes
and avoidance of severe equipment damages. Early abnormality
detection significantly reduces maintenance and lost-production
costs (Dash and Venkatasubramanian, 20 0 0). For example, large
scale syngas (gas mixture consisting primarily of hydrogen and car-
bon monoxide) production employs reformer furnaces with several
hundred catalyst-filled tubes (see the details in Section 2). Since
this process is highly integrated, a failure of a single tube can ini-
tiate a cascade of failures of several other tubes inside the furnace
box. Since the catalyst tubes in a reformer box contribute to ap-
proximately 15–20% of the total capital cost, the economic impli-
cations of early fault detection are significant. Industrial statistics
estimate the economic impact due to unplanned outages at petro-
chemical plants alone to be around $20 billion per year (Nimmo
1995). However, efficient process monitoring and control has re-
mained a challenge due to the increasing complexity of process
systems and practical limitations on continuous manual monitor-
ing. Hence, process monitoring, also commonly termed as fault
∗ ever, it is difficult and time-consuming to create an exhaustive
Corresponding author.
E-mail addresses: ankur_kumar@praxair.com, Ankur.Kumar@linde.com (A. Ku-
mar).
detection and diagnosis (FDD), has been an active field of re-
search over the past few decades (MacGregor and Kourti 1995;
Venkatasubramanian et al., 2003c; Venkatasubramanian et al.,
2003a,b, Severson et al., 2016).
FDD can be broadly classified into three categories: analyt-
ical (model-based), knowledge-based and data-driven methods
(Alzghoul et al., 2014). The analytical method relies on first-
principles-based mathematical models of the process and incor-
poration of physical understanding of the system into the fault
detection process. While the analytical models are expected to
provide superior accuracy, development of a high-fidelity math-
ematical model of complex industrial processes can be diffi-
cult. For example, modeling radiative heat transfer inside large-
scale furnace boxes in syngas plants can be quite involved
(Kumar et al., 2015). The computationally-intensive calculations
make model-based real-time monitoring of complex industrial pro-
cesses infeasible.
Knowledge-based methods are rule-based expert systems
where the rules are derived from process engineers’ experience
and plant operators’ intuitive knowledge of the underlying pro-
cess. For example, plant operators may notice that openings of
flow-control valves exhibit specific characteristic patterns before
failure; rules can be framed based on these patterns to de-
tect valve issues to prevent severe process disturbances. How-
collection of rules that covers a wide range of potential process
faults. Additionally, for novel processes, the knowledge base can

https://doi.org/10.1016/j.compchemeng.2020.106756
0098-1354/© 2020 Elsevier Ltd. All rights reserved.
2 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756

Fig. 1. Representative flowsheet of a steam methane reforming-based Hydrogen plant. Sensor measurements from the shaded region of the plant were used in process FDD
analysis. Representative process-stream temperatures adopted from (Kumar et al., 2017b). (For interpretation of the references to color in the text, the reader is referred to
the web version of this article.)

Fig. 2. Reformer Monitoring Workflow.

be extremely sparse. A good review of knowledge-based meth-
ods for fault analysis and diagnostics can be found in the work of
(Venkatasubramanian et al., 2003a).
Data-driven approach for FDD utilizes historical process data.
The data contain process information and capture intrinsic process
complexities. Hence it can be used for modeling, monitoring, and
control (Kano and Nakagawa 2008); (MacGregor et al., 2005). Pro-
cess data from faulty and normal plant operations can be used
to develop classification models that classify process conditions
into faulty and normal classes; (Yin et al., 2014b) however, large
amount of data from faulty plant operations are generally not
available to build accurate classification models. Alternatively, data
from a wide range of normal plant operations can be used to build
statistical models to determine whether the process is operating
normally. These multivariate statistical process monitoring (MSPM)
methods. (MacGregor and Kourti, 1995) have recently become
more popular due to the rapid development of process instrumen-
tation and data acquisition technology, and wide utilization of dis-
tributed control system (DCS) in modern industrial processes. Due
to the ease of accessibility to process data, and the abundance of
inexpensive data management systems, the volume of data gen-
erated from large-scale industrial processes have been on the rise
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756 3

Fig. 3. Historical data for key process variables. Note that the displayed values have been scaled by different scaling-factors for confidentiality purposes.

Fig. 4. Filtered historical data for the variables of interest for fault analysis. Note that the displayed values have been scaled by different scaling-factors for confidentiality
purposes.
4 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756
Fig. 5. Extracted steady-state and historical data for key process variables. Note that the displayed values have been scaled by different scaling-factors for confidentiality
purposes.
(Alzghoul et al., 2012). Principal component analysis (PCA) and par-
tial least squares (PLS) are among the most popular MSPM tech-
niques and rely on the projection of the high-dimensional pro-
cess data onto a lower dimensional space through latent vari-
ables for extraction of key process information (AlGhazzawi and
Lennox 20 08; Qin 20 03; Flores-Cerrillo and MacGregor 2004). Sev-
eral other data-based methods, such as independent component
analysis (ICA), artificial neural networks (ANNs), support vector
machines (SVM), kernel PCA/PLS, recursive PCA/PLS, gaussian mix-
ture models (GMMs), etc., have been explored to deal with issues
such as non-gaussianity, non-linearity, non-stationarity, and multi-
ple operating modes (Lee et al., 2004; Cho et al., 2005; Qin 1998).
Excellent reviews on the state-of-the-art data-driven FDD meth-
ods for industrial processes can be found in literature (Qin, 2012;
Ge, et al., 2013; Yin et al., 2014a; Ge, 2017; Reis and Gins 2017).
Empirical/data-driven models are especially useful in
industrial settings where low development time and payback
period, speed of implementation, and robustness to practical
issues such as missing data are important. As discussed above, a
plethora of data-driven techniques is available for the develop-
ment of a process monitoring tool. However, each method has
its advantages and shortcomings; a method that works well for
one system might not exhibit satisfactory performance for another
(Dash and Venkatasubramanian, 20 0 0; Ge et al., 2013 Ng and
Srinivasan, 2010; Perk et al., 2010). have proposed combining
multiple FDD methods in a multi-agent system for process
monitoring; these multi-agent systems, however, do not make
selection of specific FDD techniques any easier. Most of the FDD
techniques in literature are benchmarked against its performance
on the Tennessee Eastman (TE) process (Downs and Vogel 1993;
Howell et al., (1997). However, as noted by Chiang et al. (2017),
this problem is somewhat antiquated.
Very few studies are available on application of FDD methods
on real large-scale industrial systems. Paucity of publicly available
literature on demonstrated success of data-driven FDD techniques
on real large-scale industrial systems and strategies to overcome
the practical challenges (Kano and Nakagawa, 2008) (related to
model adaptation, model maintainability, data pre-processing, etc.)
faced during industrial applications are some of the reasons for
low industrial adoption of these techniques. Additionally, current
research hints towards context specificity, suggesting that model-
ers need to create appropriate combination of tools designed for
each application. Albeit, in industry, it is common to find a conser-
vative process monitoring approach where static upper and lower
alarm thresholds are used for a few key process variables, however,
as shown later in the text, this conservative approach leads to rel-
atively delayed detection of faults. Relatively faster detection is de-
pendent on plant operators serendipitously identifying these faults
from process graphs on HMI (human machine interface) screens in
the control room. Of course, this is not an ideal solution to the
fault detection problem.
In this paper, results from development and application of
an expert process monitoring system, based on data-driven
FDD methodologies, for monitoring reformer boxes in hydro-
gen plants are presented. Reformer box is a physically large
(16 m × 16 m × 12 m) scale unit operating at ~ 1800◦ F to
convert natural gas (methane) into syngas (Kumar et al., 2015).
Several methods available in literature are compared on their fault
detection capabilities. Methods for both steady-state and dynamic
process are applied and detailed step-by-step procedures have
been provided. Further, the developed expert system is successfully
tested on two separate hydrogen plants to demonstrate the sys-
tem’s replicability. The paper is organized as follows. Section 2 pro-
vides a brief overview of the hydrogen production process. Re-
former monitoring workflow is then described followed by a brief
discussion on historical process data and data pre-processing.
Sections 5 and 6 provide the details of algorithm and subsequent
results from the application of steady-state and dynamic monitor-
ing algorithms respectively.
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756
Fig. 6. Training PCA model on steady-state data: (a) Cumulative variance explained
by principal components (b) SPE statistic and 99.5% control limit (c) T2 statistic and
99.5% control limit.
2. Reformer box and hydrogen plant
In this section, we describe the reforming process in a large-
scale hydrogen plant typically producing more than a hundred mil-
lion square cubic-feet per day of high-purity hydrogen. The pur-
pose of this section is to highlight the heat-intensive, integrated
nature of the HyCO process, which justifies the utility of intelli-
gent data-driven FD technologies, for ease of replication. We begin
by describing a representative flowsheet, followed by description
of a repertoire of faults that warrant replicable, process-dependent
alarm systems, as opposed to simplistic, rule-based alarms.
Fig. 1 shows a representative flowsheet of a steam methane
reforming (SMR) based Hydrogen plant. Steam methane reform-
ing processes convert natural gas feedstock and steam into hydro-
gen. The reformer or the furnace box, has two major sections -
the radiant section (colored red in Fig. 1) and the convection sec-
tion (colored blue). The radiant section contains several burners
and catalyst-filled reformer tubes suspended vertically in a furnace
5
(Kumar et al., 2017a). The convection section consists of several
heat exchangers for waste heat recovery from the combusted flue
gas stream.
Natural gas feed (NGF) streams are used for both the reformer
feed and fuel for the burners in the furnace. The desulfurized nat-
ural gas feedstock is mixed with process steam (PS) prior to the
steam methane reformer at a specified steam to carbon (S/C) ra-
tio. This mixed feed stream enters the mixed-feed preheater in the
convection section of the reformer where it is heated by combus-
tion flue gases. The heated mixed feed stream then flows down
through multiple rows of reformer tubes within the furnace box.
The nickel-based catalyst loaded within the reformer tubes pro-
motes the following reactions
CH4 + H2 O  CO + 3H2 (1)
CO + H2 O  CO2 + H2 (2)
Reactions (1) and (2) are the steam methane reforming and
water gas shift reactions respectively. Steam reforming is a highly
endothermic reaction, favored by high temperature and low pres-
sure. Water gas shift reaction is mildly exothermic and favored
by low temperature. Both reactions are equilibrium limited with
overall conversion depending on the reformer outlet temperature
and pressure. The overall reaction is highly endothermic. The re-
quired heat is supplied by the burners through the combustion of
tail/recycle gas from the PSA unit and supplemental natural gas
fuel. Depending on the design, there are different burner arrange-
ments for a furnace. In Fig. 1, a top-fired reformer design is shown,
in which the combustion products are introduced co-current with
the process gas stream at the top of the furnace box.
The hot reformed gas exits the reformer from the lower end
of the tubes at a controlled outlet temperature around 1500 ◦ F
(Kumar et al., 2015) and is cooled against boiling water in the
process gas boiler (PGB). The reformed gas enters the high tem-
perature shift-reactor, where carbon monoxide reacts with excess
steam present in the reformed gas to produce additional hydro-
gen via the water gas shift reaction (reaction (2)). Since the wa-
ter gas shift reaction is mildly exothermic, the HTS reactor effluent
passes through a series of heat exchangers for heat recovery. The
process gas is first cooled against the natural gas feedstock within
the Natural Gas Preheater. The process gas is then used to preheat
boiler feed water in the Boiler Feedwater (BFW) Heater, and to pre-
heat deaerator makeup water in the Demineralized Water Heater.
The Steam Separator, separates process condensate from the pro-
cess gas stream which is then sent to the Pressure Swing Adsor-
bers (PSA), where final product purification takes place. The pro-
duction of a high purity hydrogen product is accomplished by the
removal of impurities (unconverted CH4 and CO) from the process
gas stream within the PSA unit.
In the convection section of the reformer, ambient air is drawn
from the atmosphere using a Forced Draft Fan (FD Fan), and heated
in the Combustion Air Preheater. The heated air and fuel mixture
are ignited in the burners located within the radiant section of the
reformer. The resultant heat of combustion is partly transferred to
the reformer tubes. A portion of the remaining heat is recovered
from the flue gas stream within the tube banks of the Mixed Feed
Preheater, the Steam Superheater, the Flue Gas Steam Generator,
and the Combustion Air Preheater. The Induced Draft Fan (ID-Fan),
assists the flow of the flue gas from the convective section to the
Flue Gas Stack, prior to discharge into the atmosphere.
The reforming process is highly integrated and fault-prone.
Large-scale hydrogen plants typically operate continuously for pro-
longed periods, usually dictated by the service contracts which re-
quire continued supply of hydrogen. The tubes experience a num-
ber of temperature and pressure cycles, the cumulative effect of
which can be extremely damaging to the tubes. The failure of one
6 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756

Fig. 7. Testing PCA model. Note that the displayed process values have been scaled by different scaling-factors for confidentiality purposes. (a) Fault imposed only in Tfg,reformer
(b) Fault imposed only in Tfg, conv (c) Fault imposed only in Tshift,out .

tube usually has a domino effect on all the tubes in the reformer,
resulting in lower hydrogen yield. In extreme cases, intense heat-
ing from abnormally high temperatures within the furnace can also
cause failure of the refractory walls of the furnace. These failures
must be detected as early as possible, such that cascading failures
can be avoided by timely planned shutdowns for repairs. Flue gas
in top-fired reformers is collected by tunnels (Kumar et al., 2015).
Maldistribution of flue gas in the tunnels can arise due to improper
designs and mechanical collapse due to fatigue which can give rise
to concentration of heat within a small region of the furnace lead-
ing to structural failures. Other common source of faults include
leakage of process gas through leaks in pipes. Leakage of high-
temperature process streams pose fire and plant operator safety
hazards. Albeit, the conventional alarm systems in the plant can
detect some of these inconsistencies, they are mostly reactive in
nature and do not attribute any deeper process knowledge.
In this work, a total of 36 variables/sensor-measurements are
used to build the monitoring expert system. Out of the 36 vari-
ables, 27 are temperature or pressure measurements of several
high-temperature process streams around and within the reformer
box. Some of these include, amongst others, temperatures of flue
gas at several locations within the radiant section, temperatures
at several locations in the convection section, and plant produc-
tion rate. Note that monitoring the high-temperature regions of the
plant was of prime importance and therefore temperatures of rel-
atively colder process streams are not included. For confidentiality
reasons, the locations of all the output sensors cannot be disclosed.
The rest of the variables include fuel flow, PSA-recycle flow, and
combustion air flow into the burners, process feed and steam flow
into the reformer tubes, heat content of natural gas, and tempera-
tures of a few reformer input streams.
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756
Fig. 8. Steps involved in external analysis method.
3. Reformer monitoring workflow
Fig. 2 shows the workflow for real-time reformer-unit monitor-
ing system. The workflow executes at regular intervals on a sched-
ule. During each run, recent process data are retrieved from his-
torian and pre-processed to filter out measurement noise and bad
Fig. 9. Training external-analysis model on steady-state data: (a) Cumulative vari-
ance explained by principal components (b) SPE statistic and 99.5% control limit (c)
T2 statistic and 99.5% control limit.
7
measurements. Pre-processed data is analyzed by the monitoring
algorithm and the computed process metrics are compared against
threshold values. Upon successful fault detection, the faulty vari-
ables are identified by the fault-diagnosis block and relevant tex-
tual and graphical information are displayed on an operator in-
terface. On the interface, plant operators can confirm the accu-
racy of the faults, acknowledge the true alarms, and reject the
false alarms. The feedback from plant operators on fault accuracy
is used to update the model. In the following sections, each com-
ponent of the workflow is described in detail.
4. Plant historical data and pre-processing
About 3 years of data, sampled every minute, consisting about
1.4 million measurement samples, have been used for the develop-
ment of the fault-detection model. A moving-average filter with 30
min (30 samples) window is used to remove measurement noise
in the raw historian data. Fig. 3 shows the filtered historical data
for some key reformer input variables. Significant variations in val-
ues of the input variables have been captured in the data. Plant
operation state changes due to changes in product demand, am-
bient temperature, etc. Such richness in data ensures that the de-
veloped model has been trained with a representative plant-input
space which encompasses the typical normal variations in plant
operation state. Such practice consequently, keeps the frequency
of false fault-alarms low. In Fig. 3 the large drops in process val-
ues (where values drop below the displayed axes) correspond to
plant shutdown periods or invalid historian data (resulting from
sensor issues or bad communication between plant control system
and historian). Invalid data and data from the shutdown periods
are removed from the training data prior to model development.
Additionally, a few hours of data before and after the shutdown
periods are removed as well. Plant shutdown is identified by com-
paring fuel flow to burner to a threshold value: fuel flow lower
than threshold signals plant shutdown. Removal of outliers is an-
other important pre-processing step where data with uncharacter-
istic high or low values are removed. Presence of outliers can de-
crease the sensitivity of monitoring system and cause delay in de-
tection of faults. Any value for a variable that is beyond five stan-
dard deviations from the mean for that variable has been treated
as an outlier in this work.
The historical period did not contain any instances where actual
process fault had occurred. However, if the fault-instances are only
a few in number then they need not be explicitly identified and re-
moved from the training period; since the length of faulty periods
would be much smaller than that of normal operations, the pres-
ence of fault-instances in training data would not impact model-
development significantly. True faults are usually accompanied by
plant shutdowns. During pre-processing, data in the vicinity of the
shutdown event are filtered out. In unusual cases, where there are
high volumes of faulty data, appropriate filters need to be utilized.
The non-requirement of excluding occasional faulty data from the
training data serves an important practical utility: it allows the
8 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756

Fig. 10. Testing external-analysis model. Note that the displayed process values have been scaled by different scaling-factors for confidentiality purposes. (a) Fault imposed
only in Tfg,reformer (b) Fault imposed only in Tfg, conv (c) Fault imposed only in Tshift,out .

work process for replication of the tool to other sites within the
organization simple and convenient.
4.1. Variables of interest for fault analysis
In this work, 3 of the output variables - temperature of flue gas
from reformer (Tfg,reformer ), temperature of flue gas near the end of
convection section (Tfg, conv ), and shift-reactor outlet temperature
(Tshift,out ) - are chosen for fault analysis. The choice is arbitrary and
any other monitored variable could also have been chosen to illus-
trate the technology discussed in this article. To mimic the fault
scenarios generally observed during plant operations, a simulated
fault manifested by increase in temperature values have been in-
troduced in these variables. Fig. 4 shows the historical data for
these variables; all the variables display considerable variations in
operating values. A common approach, typically followed in pro-
cess industry for abnormality detection in these variables, is to put
upper and lower alarm limits spanning the observed range of val-
ues. This approach, however, is inadequate since several potential
faults where the faulty variable remains within the normal oper-
ating range can remain undetected. This is primarily due to the
multivariate nature of the process. Moreover, the attribution of the
normal operating range is based upon specific operating conditions
that can change due to fluctuations in demand. In such cases, ei-
ther the normal operating range has to be wide spanning multi-
ple regimes, or the operating regimes need to be reset every time
a change in operating conditions is encountered. In the following
sections FD methodologies are investigated that overcome the lim-
itations of this simple approach for reformer monitoring.
5. Steady-state monitoring algorithm
Steady-state monitoring algorithms train fault-models using
steady-state training data to identify normal operating states. Dur-
ing model-training, control limits on one or more fault detection
indices are determined. During testing, faulty/abnormal process
data that deviates from the normal process behavior and conse-
quently, violate the control limits are flagged as faults.
5.1. Steady-state extraction
For monitoring based on steady-state data, steady-state periods
need to be extracted out of the historical time-series data which
is then used as the training data-set. The approach by Simon and
Litt (2011) for identification of steady-state operating points has
been adopted in this work. An overview of the methodology is
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756 9

Fig. 11. Normal dynamic changes in key process variables. Note that the displayed values have been scaled by different scaling-factors for confidentiality purposes.

Fig. 12. Normal dynamic changes in Tfg,reformer , Tfg, conv , and Tshift,out . Note that the displayed values have been scaled by different scaling-factors for confidentiality purposes.

Algorithm 1 Steady-state periods extraction.

1: Calculate standard deviation (σo) of key variables from

sample steady-state data
2: Assign: steadyStateflagi = 0 ∀i ∈ 1 to N
N is number of measurement samples
3: for i = 1 to N − m + 1 do {m is number of samples in
test-window}
4: Assign data samples from i to i + m − 1 to test-
window
5: if all measurements are valid & there is no plant shut-
down in test-window then
6: Calculate standard deviation (σ ) of key variables
from test-window data
7: if σ < 2σo ∀ key variables then
8: steadyStateflag j = 1 ∀ j ∈ i to i + m − 1
9: end if Fig. 13. Q statistic from steady-state external-analysis model for data during nor-
mal process changes.
10: end if
11: end for
12: Samples with steadyStateflagi = 1, where i is sample
number, belong to steady-state operation period
10 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756

Fig. 14. Augmenting a matrix with lagged measurements (yij refers to the value of jth variable from the ith measurement sample). Note that if any value in a row in the
augmented input matrix or output matrix is invalid (due to bad measurements or sensor issues), then the entire row from both the augmented input matrix and output
matrix is removed.

Fig. 15. Steps involved in dynamic external analysis method.

outlined in Algorithm 1. A 5h sample (filtered) data-set is used to
calculate nominal standard deviations of a key reformer variables,
such as fuel flow to burners, reformed-gas temperature, etc. Sam-
ples belonging to steady-state operation periods are identified by
comparing the respective standard deviations calculated in a slid-
ing test-window of size 5h ( ~ 300 samples), with nominal values.
Fig. 5 shows the extracted steady-state operation periods ob-
tained from application of Algorithm 1; about 0.7 million sam-
ples are retained. It can be observed that operation periods
with significant increase or decrease in process values have been
excluded.
5.2. PCA-based monitoring
PCA models are primarily used to extract correlations among
process variables from data. Abnormal/faulty process data do not
follow normal process correlations and violate one or more fault
detection indices. While the reader is referred to the work by
Qin (2012) for the underlying mathematical details, step-by-step
procedure for practical application of the method is provided in
Algorithm 3 in Appendix section (A.1). The fault detection indices
used in PCA are T2 and Q statistics. The Q statistic, defined as the
sum of squared residuals or the sum of prediction errors (SPE),
measures abnormality in data that breaks the normal process cor-
relations. The T2 statistic measures the distance of the projec-
tion of the sample from the origin, in the principal component
subspace. It is a metric that measures the variance within the
model.
5.2.1. Training model & application
Fig. 6a shows the cumulative variance in training data captured
by the principal components (PCs). 12 PCs capturing 99% variance
were retained. Fig. 6b and c show the resulting Q and T2 statistics
for the training data; the control limits of 2.8 and 84 are deter-
mined such that the number of samples outside the control limits
is only 0.5% of the entire training samples (Yamamoto et al., 2004;
Kano et al., 2004). For gauging the fault-detection capability of the
developed model, the 5-hour sample steady state operation data
that was used in Fig. 5 is considered and 5% faults are imposed
in Tfg,reformer , Tfg, conv , and Tshift,out , separately in 3 different simu-
lations, as shown in Fig. 7. The fault in each of the 3 variables is
introduced at the start of the 5th h and is linearly increased till the
end of the hour. It can be seen that while the fault in Tshift,out leads
to breach of SPE control limit, faults in Tfg,reformer and Tfg, conv go
undetected. This method has been discussed in this paper despite
inadequate monitoring performance so that it serves as a reference
while assessing the performance of the methodologies discussed in
the following sections.
5.3. External analysis
Kano et al. (2004) proposed external analysis for multivariate
statistical process monitoring to explicitly take into account the
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756 11

change in operating conditions of ‘main’ or output variables due

to changes in ‘external variables’ or input variables. Fig. 8 gives
an overview of the method and Algorithm 2 provides detailed
step-by-step procedure for this methodology. A (PLS) model be-
tween main variables (reformer outputs) and external variables
(reformer inputs) is developed to decompose main variables into
two parts: one part is explained by external variables and the other
part is not explained. PCA is then performed on the unexplained
part (output residuals) and a (99.5%) control limit for SPE is de-
termined. Note that if the input variables are uncorrelated then
ordinary least squares (OLS) can be used in place of PLS. How-
ever, in industrial settings, input variables tend to be correlated
and PLS is well-suited to handle such correlations. If the rela-

Algorithm 2 Steady-state external analysis.

1: Training model:
2: Generate output matrix (Y ∈ N×q ) and input matrix
(X ∈ N×p ) where N, q, p are numbers of measurement
samples, output variables, and input variables, respec-
tively
3: Calculate mean (y ∈ 1×q , x ∈ 1×p ) and standard de-
viation (σy ∈ 1×q , σx ∈ 1×p ) vectors: ith element of y
is mean of ith column of Y . Calculate similarly for other
vectors.
4: Normalize Y , X to Y  ∈ N×q , X  ∈ N×p : ith element of
each row of Y is subtracted by ith element of y and then
divided by ith element of σy . Calculate similarly for X  Fig. 16. Training dynamic external-analysis model: (a) SPE statistic and 99.5% con-

5:  
Perform PLS regression with X and Y . Choose number
trol limit (b) SPE statistic and 99.5% control limit.

of latent variables such that 99% variance in input vari-

ables is captured. PLS regression gives coefficient matrix
β ∈  p×q tionship between the main and external variables is highly non-
6: Reconstruct: Y∗ = X  × β . De-normalize Y ∗ to Y ∗ . ith linear then nonlinear extension of PLS (kernel PLS) (Qin, 2012),
element of each row of Y ∗ is added by ith element of neural networks, or any other suitable nonlinear modeling tech-
y and then multiplied by ith element of σy . nique (Ge et al., 2008) may be used for obtaining the output
7: Compute output residuals: Yr = Y − Y ∗ residuals.
8: Normalize Yr to Y r . Let y , σy be the mean and stan- Some researchers have argued that the use of SPE is more
r r
dard deviation vectors. preferable than the T2 statistic (a normal change in the operating
9: Perform PCA on Y r . PCA generates: conditions can potentially violate the control limits for T2 statistic)
– Score matrix T ∈ N×q for fault detection Qin (2003), Yamamoto et al. (2004). However,
– Coefficient matrix C ∈ q×q : ith column of C con- in the external analysis method, the T2 statistic is an equally im-
tains coefficients for the ith principal component (PC) portant metric. For example, consider a case where the residual
in variable ‘A’ is uncorrelated to the rest of the residuals. In the
10: Select number of PCs (ncomp ) such that 99% variance
r is captured. presence of an abnormal increase in the value of only the variable
in Y ‘A’, its residual will show an abnormal increase. However, due to
∗ = Tn
comp ∗ (Cncomp )
transpose
11: Reconstruct: Y r : Tncomp , Cncomp no correlation, the residual of variable ‘A’ will get reconstructed
are the first ncomp columns of T and C, respectively. (in step 20 in Algorithm 2) with abnormally high value and the Q
12: Calculate error matrix: E = Y r − Y
∗ value would remain small. On the other hand, the T2 statistic will
r
13: Compute Q ∈ N×1 vector: ith element of Q is dot- show an spike and can potentially breach its control limit and flag
product of ith row of E matrix a fault.
14: Plot Q and find control limit QCL such that only 0.5%
of Qs are above QCL 5.3.1. Training & application
6 latent variables, determined via cross-validation, capturing
15: Testing model:
78.4% variance in output variables were used in PLS regression.
16: Let y ∈ 1×q and x ∈ 1×p be the output and input test Fig. 9a shows the cumulative variance in output residuals captured
vectors by the principal components. 15 PCs capturing 99% variance were
17: Compute  y = (y − y )/σy and  x = (x − x )/σx : opera- retained. Fig. 9b and c shows the resulting Q and T2 statistics for
tions are performed element-wise. the training data; the control limit of 2.08 and 89 are again de-
18: Calculate test-output residual (yr ):  x × β ; y∗ =
y∗ =  termined such that the number of samples outside the control
y∗ ∗ σy + y; yr = y − y∗
 limit is only 0.5% of the entire training samples. Again, for gauging
19: Compute  yr = (yr − yr )/σyr the fault-detection capability of the developed model, the 5-hour
20: Calculate  yr ∗ Cncomp ∗ (Cncomp )transpose
y∗r =  steady state operation data that was used in Fig. 5 is considered
21: Compute Q =  y∗r 2
yr −  and 5% faults were imposed. These faults were easily detected for
all 3 of Tfg,reformer , Tfg, conv , and Tshift,out , and flagged. Fig. 10 shows
12 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756

Fig. 17. Testing dynamic external-analysis model. Note that the displayed process values have been scaled by different scaling-factors for confidentiality purposes. (a) Fault
imposed only in Tfg,reformer (b) Fault imposed only in Tfg, conv (c) Fault imposed only in Tshift,out (d) Fault imposed in Tfg,reformer and Tfg, conv .
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756
Fig. 18. Testing dynamic external-analysis model for data during normal process
changes.
the least fault magnitudes that cause SPE to breach the control
limit. It can be seen that only ~ 2% faults cause the breach.
Note that only Q plots are shown because the T2 values did not
show appreciable increase. Thus, the fault sensitivity or the fault-
detection capability of steady-state external analysis for reformer
unit is superior to PCA-based approach.
A problem associated with training models with steady-state
operation data is that such models can be used for fault detec-
tion only when the input variables are steady. While it seems rea-
sonable to selectively monitor a process for faults only when in-
puts are steady, fault scenarios where abnormality in process out-
put leads to changes in process inputs (for example, unexpected
changes in flue gas temperature can affect the reformed gas tem-
perature and the control system will manipulate fuel flow to burn-
ers to keep the reformed gas temperature steady) will not be mon-
itored and the faults will remain undetected. When the inputs are
undergoing normal change and the whole plant is moving from
one operation state to another, steady-state models, if used, can
report high percentage of false fault-alarms; this happens because
process variables do not maintain steady-state correlations during
the transient phase when moving from one steady-state condi-
tion to the other, leading to high SPE values. Figs. 11, 12 shows
one such instance where the plant experiences normal dynamic
changes leading to ~ 10% variations in output temperatures within
a few hours. Q statistic, however, flags this normal transient pe-
riod as faulty period (Fig. 13). The inability of this method to han-
dle normal process transients makes it unsuitable to be used as
a stand-alone monitoring algorithm. An approach to combine this
method with other algorithms is discussed in Section 7.
13
6. Dynamic monitoring algorithm
For a dynamic system, the current values of a system depend on
the past values. For the reformer, it has been shown that the oper-
ating conditions vary frequently according to production require-
ments. Therefore, it is important that the temporal correlations
among reformer variables are also taken into account during devel-
opment of the fault-model; this enables the model to be used for
fault-detection irrespective of whether the reformer is at steady-
state or not. A convenient data-based scheme with low compu-
tation cost for statistical process monitoring for dynamic systems
was proposed by Ku et al. (1995) for PCA analysis. In their ap-
proach, past or lagged measurements are treated as additional pro-
cess variables and are utilized to build an augmented data matrix.
Traditional SPM analysis is performed on this augmented matrix
for model development. Ku et al. (1995) and Tsung (20 0 0) used
dynamic PCA to recover the autoregressive structure from process
data. Several studies have been reported in literature on successful
implementation of data augmentation for modeling dynamic pro-
cesses. De Ketelaere et al. (2015), Dong and Qin (2018b), Dong and
Qin (2018c), Dong and Qin (2018a), Chen and McAvoy (1998).
Fig. 14 shows the procedure for building an augmented matrix
with l − 1 lagged variables; yij refers to the value of jth variable
from the ith measurement sample. For PLS modeling, a straight-
forward analogous approach is to include lagged values of input
variables to generate an augmented input matrix and perform tra-
ditional PLS analysis on output matrix and the augmented input
matrix. Ricker (1988) For reformer monitoring, dynamic external
analysis had promising performance and is presented in the fol-
lowing subsection.
6.1. Dynamic external analysis
Kano et al. (2004) proposed a dynamic extension of external
analysis method. The procedure (Algorithm 4 in Appendix) is sim-
ilar to that of steady-state external analysis except that the input
data matrix is replaced by its augmented version. Fig. 15 gives an
overview of the method. Dynamic PLS is followed by a PCA on out-
put residuals and the 99.5% control limits, QCL and TCL 2 are deter-
mined. Based on the typical time taken by a reformer to settle to
a new steady-state upon step-changes in inputs, l = 15 was em-
ployed for data augmentation. Model development was performed
in MATLAB (MAT, 2016b) on a Windows PC with 16 GB RAM.
6.1.1. Testing & application
15 latent variables capturing 71.4% variance in output variables
were used in dynamic PLS regression. 16 PCs capturing 99% vari-
ance were retained. Fig. 16 shows the resulting Q and T2 statis-
tics after logarithmic transformation for the training data; a control
limit of 3.35 and 92 are obtained. Fig. 17 shows the least fault mag-
nitudes that cause SPE to breach the control limit. ~ 3% faults in
Tfg,reformer and Tfg, conv , and ~ 5% fault in Tshift,out cause the breach.
Thus, the fault sensitivity, slightly worse than that of steady-state
external analysis method, is adequate for reformer unit monitor-
ing. Fig. 17d shows the error metric plot when faults in multiple
variables were imposed; only 1.5% faults cause control limit vio-
lation. This highlights a crucial advantage of the multivariate sta-
tistical analysis over univariate analysis or the methods that rely
on individual variable thresholds/alarm limits for fault detection.
In multivariate analysis, faults of smaller magnitudes occurring si-
multaneously in multiple variables can compound to cause alarm
threshold breach.
The primary motivation for using dynamic variation of the ex-
ternal analysis method was to avoid false alarms during normal
process transients. Fig. 18 gives the plots of the fault-detection
metrics for the normal process data from Fig. 11. It can be observed
14 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756

Fig. 19. Contribution plots from testing dynamic external-analysis model. (a) Fault imposed only in Tfg,reformer (b) Fault imposed only in Tfg, conv (c) Fault imposed only in
Tshift,out (d) Fault imposed in Tfg,reformer and Tfg, conv .

Fig. 20. Monitoring workflow based on dynamic and steady-state external analysis.

that, unlike steady-state external analysis, dynamic external analy- from each output variable. Qin (2003) For example, step 13 in
sis results in Q (and T2 ) values that are well below the threshold Algorithm 4 can be decomposed as follows:
value. Note that this period of process data was removed from the

q
training data-set. SP E =  y∗r 2 =
yr −  SP Ei (3)
i=1

6.1.2. Fault diagnosis

Once the presence of a fault has been detected, the next step SP Ei = ( y∗r, j )2
yr, j −  (4)
is to find the faulty output variables. Contribution plots are the
popular diagnostic tools for fault identification and are gener- In Eq. (4), 
yr, j refers to the residual in the jth variable. Vari-
ated by breaking down the SPE metric value into contributions ables corresponding to the largest SPEs are considered major con-
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756
Fig. 21. Historical data for process variables form another the second hydrogen manufacturing site. Note that the displayed values have been scaled by different scaling-
factors for confidentiality purposes.
Fig. 22. Training dynamic external-analysis model for the second hydrogen manu-
facturing site: SPE statistic and 99.5% control limit.
tributors to the fault. A known shortcoming of this approach is
’smearing’ effect, where contribution from one variable is propa-
gated to other variables during calculation of the PCA projections
onto latent directions; this, correspondingly, can result in mis-
diagnosis Qin (2003). An alternate approach, reconstruction-based
contribution, has been proposed by Alcala and Qin (2009) where
15
the contribution of a variable is defined as the amount of ‘re-
construction’ along the variable direction that minimizes the
fault-detection index. Alcala and Qin (2011) further demonstrated
the superiority of the method over the traditional contribution
plot diagnosis method. The reader is referred to Alcala and
Qin (2009) for mathematical details of the method. Fig. 19 shows
the reconstruction-based contribution plots for the Q statistic for
faults shown in Fig. 17. Note that variable numbers assigned to
Tfg,reformer , Tfg, conv , and Tshift,out are 5, 13, and 27, respectively. For
all the fault scenarios, correct faulty variables have been identi-
fied. The ease of interpretability of fault diagnostics through con-
tribution plots is one of the major advantages of PCA/PLS-based
approaches.
7. Monitoring algorithm workflow
Post-deployment of the process monitoring tool, false positives
(fault alarm without actual fault) and false negative (no fault alarm
when there are actual faults) are both undesirable; while the for-
mer leads to a loss of user’s confidence in the tool, the latter
leads to delayed fault detection by the plant operators. In pre-
vious sections, it was observed that while steady-state external
16 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756

Fig. 23. Testing dynamic external-analysis model at the second hydrogen manufacturing site. Note that the displayed process values have been scaled by different scaling-
factors for confidentiality purposes. (a) Fault imposed only in Tfg,reformer (b) Fault imposed only in Tfg, conv (c) Fault imposed only in Tshift,out .

analysis can lead to false positives during normal process tran-
sients, dynamic external analysis can lead to false negatives due
to lower fault sensitivity. To combine the benefits of both these
approaches, the workflow depicted in Fig. 20 serves as the mon-
itoring algorithm for the reformer unit. Depending upon whether
the inputs are at steady state or not, fault-detection is performed
using steady-state or dynamic external analysis. For steady-state
determination, a methodology similar to that in Algorithm 1 is
utilized. Moreover, to avoid false positives from point fluctu-
ations in sensor measurements, Q and T2 are required to be
above QCL and TCL2 , respectively, continuously for some period of
time.
8. FD application at identical plant
An advantage offered by the data-based monitoring method-
ologies is that once a method has been found that gives satis-
factory performance for a particular system, the method can po-
tentially be applied without many modifications for monitoring
other similar systems. This keeps the investment of labor, time,
and money for model development during tool replication low. For
reformer monitoring, the external analysis-based method was re-
peated for monitoring another identical reformer-based hydrogen
manufacturing plant. Satisfactory fault-detection performance was
again obtained. Fig. 21 shows the filtered historical data (sampled
every minute over 3 years) from the other site for the key re-
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756 17

former variables along with the fault-relevant variables. Consider- ogy. A combination of steady-state external analysis and dynamic
able variations can be noticed in the values of all the process vari- external analysis is finally proposed to overcome the trade-off
ables. Fig. 22 shows the Q plot for the training data. Fault detection between false-positives (from steady-state external analysis) and
tests (Fig. 23) confirms good fault sensitivity where 1.5%, 3%, and false-negatives (from dynamic external analysis). Several aspects
4% faults in Tfg,reformer , Tfg, conv , and Tshift,out , respectively, cause QCL which affect the adoption and success of any analytic product
breach. within an organization have been highlighted: these aspects in-
clude ease of replication, time to replicate, frequency of false pos-
9. Discussion and recommendations itives and negatives, interpretability of results from the tool. The
detailed description of application of data-driven process moni-
In the previous sections, it was shown how process relation- toring methodologies for large-scale systems is expected to guide
ships can be extracted from process data to build process and other industrial practitioners and facilitate wider industrial adop-
fault-models for reformer boxes quickly. While this is very con- tion.
venient, model-developers (and the end-users) should be aware
of the limitations of data-driven approaches. If the historical pro- Declaration of Competing Interest
cess data, for some reason, do not accurately represent the be-
havior of the current system, frequency of false alarms will in- The authors declare that they have no known competing finan-
crease significantly. For example, in the hydrogen plant, cata- cial interests or personal relationships that could have appeared to
lyst activity decreases over time which affects the temperature influence the work reported in this paper.
of several process streams. Equipment, such as heat exchangers,
may be replaced causing step-changes in variable relationships Appendix A
around the equipment which impact the overall process correla-
tions. Methodologies should be in place to deal with such sce- A1. PCA-based monitoring
narios. Since the catalyst deactivation is a slow change, it can be
handled by re-training the model at regular intervals and provid-
ing higher weightage to latest data. Step changes in process cor- Algorithm 3 PCA model.
relations may necessitate taking the monitoring tool offline un-
til enough data have been gathered to re-develop the model. The 1: Training model:
monitoring system presented in this work identifies the need for 2: Generate data matrix (Z ∈ N×(q+ p) ) where N, q, p
model update by tracking the user feedback on false positives. If are numbers of measurement samples, output variables,
frequency of false positives is high, the model is re-trained us- and input variables, respectively
ing more recent data. However, if false positives are isolated oc- 3: Calculate mean (z ∈ 1×(q+ p) ) and standard deviation
currences, then a bank of false positive scenarios may be main-
(σz ∈ 1×(q+ p) ) vectors: ith element of z is mean of ith
tained which can be used to identify and avoid repeating false
column of Z. Calculate similarly for standard deviation.
alarm.
4: Normalize Z to  Z ∈ N×(q+ p) : ith element of each row
Another important consideration during model development
of Z is subtracted by ith element of z and then divided
is choosing values for several model parameters. These param-
eters include number of lags used for data augmentation (l),
by ith element of σz .
number of latent components to retain in PCA and PLS, control 5: Perform PCA on  Z . PCA generates:
limit for SPE (QCL ). Several studies have been published that pro- – Score matrix T ∈ N×(q+ p)
vide guidance on proper selection of values for these parameters. – Coefficient matrix C ∈ (q+ p)×(q+ p) : ith column of C
Kano et al. (2004) used process time-constant to determine the lag contains coefficients for the ith principal component
period (l) for a CSTR process. The number of components in PCA (PC)
and PLS can be determined using a scree plot, cross validation, or 6: Select number of PCs (ncomp ) such that 99% variance
the cumulative percentage of variance De Ketelaere et al. (2015), in 
Z is captured.
Li et al. (20 0 0). An alternate approach is to collectively determine 7: Reconstruct:  Z ∗ = Tncomp ∗ (Cncomp )transpose : Tncomp , Cncomp
these parameters via automated trial and error; a minimum fault- are the first ncomp columns of T and C, respectively.
sensitivity criteria can be set and the parameter space can be ex- 8: Calculate error matrix: E =  Z −Y Z ∗
haustively searched for the values that meet the specified criteria.
9: Compute Q ∈ N×1 vector: ith element of Q is dot-
product of ith row of E matrix
10. Conclusion
10: Plot Q and find control limit QCL such that only 0.5%
of Qs are above QCL
In this work, development and application of process data-
based process monitoring has been reported for a large-scale 11: Testing model:
reformer-box unit of a hydrogen manufacturing plant. A complete
expert system workflow, from retrieving data from data historian 12: Let z ∈ 1×(q+ p) be the test measurement vector.
to displaying fault details on plant operator’s screen, has been 13: Compute  z = (z − z )/σy : operations are performed
provided. External analysis was found to provide the best fault- element-wise.
detection performance. Similar fault-detection performance was 14: Calculate 
z∗ = z ∗ Cncomp ∗ (Cncomp )transpose
obtained during direct application of the method at another hy- 15: Compute Q =  z −z ∗ 2
drogen plant with little modifications in the overall methodol-
18 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756

A2. Dynamic external analysis References

Algorithm 4 Dynamic external analysis.
1: Training model:
2: Generate output matrix (Y ∈ (N−l+1)×q ) and aug-
mented input matrix (X ∈ (N−l+1)×pl ) where N, q, p are
numbers of augmented measurement samples, output
variables, and input variables, respectively, and l is the
lag-period.
3: Calculate mean (y ∈ 1×ql , x ∈ 1×pl ) and standard de-
viation (σy ∈ 1×ql , σx ∈ 1×pl ) vectors: ith element of y
is mean of ith column of Y . Calculate similarly for other
vectors.
Alcala, C.F., Qin, S.J., 2009. Reconstruction-based contribution for process monitor-
ing. Automatica 45 (7), 1593–1600.
Alcala, C.F., Qin, S.J., 2011. Analysis and generalization of fault diagnosis methods for
process monitoring. J. Process Control 21 (3), 322–330.
AlGhazzawi, A., Lennox, B., 2008. Monitoring a complex refining process using mul-
tivariate statistics. Control Eng. Pract. 16 (3), 294–307.
Alzghoul, A., Backe, B., Löfstrand, M., Byström, A., Liljedahl, B., 2014. Comparing a
knowledge-based and a data-driven method in querying data streams for sys-
tem fault detection: a hydraulic drive system application. Comput. Ind. 65 (8),
1126–1135. doi:10.1016/j.compind.2014.06.003.
Alzghoul, A., Löfstrand, M., Backe, B., 2012. Data stream forecasting for system fault
prediction. Comput. Ind. Eng. 62 (4), 972–978. doi:10.1016/j.cie.2011.12.023.
Chen, G., McAvoy, T.J., 1998. Predictive on-line monitoring of continuous processes.
J. Process Control 8 (5–6), 409–420.
Chiang, L., Lu, B., Castillo, I., 2017. Big data analytics in chemical en-
gineering. Annu. Rev. Chem. Biomol. Eng. 8 (1), 63–85. doi:10.1146/
annurev- chembioeng- 060816- 101555.
Cho, J.-H., Lee, J.-M., Choi, S.W., Lee, D., Lee, I.-B., 2005. Fault identification for pro-

4: Normalize Y , X to Y  ∈ N×ql , X ∈ N×pl : ith element of
each row of Y is subtracted by ith element of y and then
divided by ith element of σy . Calculate similarly for X
5:  
Perform PLS regression with X and Y . Choose number
of latent variables such that 99% variance in input vari-
ables is captured. PLS regression gives coefficient matrix
β ∈  pl×ql
6: Reconstruct: Y∗ = X  × β . De-normalize Y ∗ to Y ∗ . ith
 ∗
element of each row of Y is added by ith element of
y and then multiplied by ith element of σy .
7: Compute output residuals: Yr = Y − Y ∗
8: Normalize Yr to Y r . Let y , σy be the mean and stan-
r r
dard deviation vectors.
cess monitoring using kernel principal component analysis. Chem. Eng. Sci. 60
(1), 279–288.
Dash, S., Venkatasubramanian, V., 20 0 0. Challenges in the industrial applications of
 fault diagnostic systems. Comput. Chem. Eng. 24 (2–7), 785–791.
De Ketelaere, B., Hubert, M., Schmitt, E., 2015. Overview of PCA-based statistical pro-
cess-monitoring methods for time-dependent, high-dimensional data. J. Qual.
Technol. 47 (4), 318–335.
Dong, Y., Qin, S.J., 2018. Dynamic latent variable analytics for process operations and
control. Comput. Chem. Eng. 114, 69–80.
Dong, Y., Qin, S.J., 2018. A novel dynamic PCA algorithm for dynamic data modeling
and process monitoring. J. Process Control 67, 1–11.
Dong, Y., Qin, S.J., 2018. Regression on dynamic PLS structures for supervised learn-
ing of dynamic data. J. Process Control 68, 64–72.
Downs, J.J., Vogel, E., 1993. A plant-wide industrial process control problem. Com-
put. Chem. Eng. 17 (3), 245–255. doi:10.1016/0 098-1354(93)80 018-I.
Flores-Cerrillo, J., MacGregor, J.F., 2004. Multivariate monitoring of batch processes
using batch-to-batch information. AlChE J. 50 (6), 1219–1228.
Ge, Z., 2017. Review on data-driven modeling and monitoring for plant-wide indus-

9: Perform PCA on Y r . PCA generates: trial processes. Chemometr. Intell. Lab. Syst. 171, 16–25.
Ge, Z., Song, Z., Gao, F., 2013. Review of recent research on data-based process mon-
– Score matrix T ∈ N×ql itoring. Ind. Eng. Chem. Res. 52 (10), 3543–3562. doi:10.1021/ie302069q.
Ge, Z., Yang, C., Song, Z., Wang, H., 2008. Robust online monitoring for multimode
– Coefficient matrix C ∈ ql×ql : ith column of C con- processes based on nonlinear external analysis. Ind. Eng. Chem. Res. 47 (14),
tains coefficients for the ith principal component (PC) 4775–4783.
10: Select number of PCs (ncomp ) such that 99% variance Howell, J., Chen, J., Zhang, J., 1997. The tennessee eastman problem as a pro-
r is captured.
in Y
cess monitoring benchmark. In: Proceedings of IFAC Symposium Safeprocess’97,
pp. 222–227. doi:10.1016/S1474- 6670(17)42405- 0.
∗ = Tn
comp ∗ (Cncomp )
transpose
11: Reconstruct: Y r : Tncomp , Cncomp Kano, M., Hasebe, S., Hashimoto, I., Ohno, H., 2004. Evolution of multivariate statis-
tical process control: application of independent component analysis and exter-
are the first ncomp columns of T and C, respectively. nal analysis. Comput. Chem. Eng. 28 (6–7), 1157–1166.
12: Calculate error matrix: E = Y r − Y
∗ Kano, M., Nakagawa, Y., 2008. Data-based process monitoring, process control, and
r
13: Compute Q ∈  N×1
vector: ith element of Q is dot- quality improvement: recent developments and applications in steel industry.
Comput. Chem. Eng. 32 (1–2), 12–24.
product of ith row of E matrix Ku, W., Storer, R.H., Georgakis, C., 1995. Disturbance detection and isolation by
14: Plot Q and find control limit QCL such that only 0.5% dynamic principal component analysis. Chemometr. Intell. Lab. Syst. 30 (1),
179–196.
of Qs are above QCL Kumar, A., Baldea, M., Edgar, T.F., 2017. A physics-based model for industrial
steam-methane reformer optimization with non-uniform temperature field.
15: Testing model: Comput. Chem. Eng. 105, 224–236.
Kumar, A., Baldea, M., Edgar, T.F., Ezekoye, O.A., 2015. Smart manufacturing approach
16: Let y ∈ 1×ql and x ∈ 1×pl be the augmented output for efficient operation of industrial steam-methane reformers. Ind. Eng. Chem.
and input test vectors Res. 54 (16), 4360–4370.
17: Compute  y = (y − y )/σy and  x = (x − x )/σx : opera- Kumar, A., Edgar, T.F., Baldea, M., 2017. Multi-resolution model of an indus-
trial hydrogen plant for plantwide operational optimization with non-uniform
tions are performed element-wise. steam-methane reformer temperature field. Comput. Chem. Eng. 107, 271–283.
18: Calculate test-output residual (yr ):  x × β ; y∗ =
y∗ =  Lee, J.-M., Yoo, C., Lee, I.-B., 2004. Statistical process monitoring with independent
y ∗ σy + y; yr = y − y∗
 ∗ component analysis. J. Process Control 14 (5), 467–485.
Li, W., Yue, H.H., Valle-Cervantes, S., Qin, S.J., 20 0 0. Recursive PCA for adaptive pro-
19: Compute  yr = (yr − yr )/σyr cess monitoring. J. Process Control 10 (5), 471–486.
20: Calculate  yr ∗ Cncomp ∗ (Cncomp )transpose
y∗r =  MacGregor, J.F., Kourti, T., 1995. Statistical process control of multivariate processes.
Control Eng. Pract. 3 (3), 403–414.
21: Compute Q =  y∗r 2
yr −  MacGregor, J.F., Yu, H., Muñoz, S.G., Flores-Cerrillo, J., 2005. Data-based latent vari-
able methods for process analysis, monitoring and control. Comput. Chem. Eng.
29 (6), 1217–1223.
MATLAB and Statistics Toolbox, 2016b. The MathWorks, Inc., Natick, Massachusetts,
United States.
Ng, Y.S., Srinivasan, R., 2010. Multi-agent based collaborative fault detection and
CRediT authorship contribution statement identification in chemical processes. Eng. Appl. Artif. Intell. 23 (6), 934–949.
Nimmo, I., 1995. Adequately address abnormal operations. Chem. Eng. Prog. 91 (9).
Perk, S., Teymour, F., Cinar, A., 2010. Statistical monitoring of complex chemical pro-
Ankur Kumar: Conceptualization, Methodology, Software, For- cesses using agent-based systems. Ind. Eng. Chem. Res. 49 (11), 5080–5093.
mal analysis, Data curation, Writing - original draft, Visualiza- Qin, S.J., 1998. Recursive PLS algorithms for adaptive data modeling. Comput. Chem.
Eng. 22 (4–5), 503–514.
tion, Project administration. Apratim Bhattacharya: Conceptualiza-
Qin, S.J., 2003. Statistical process monitoring: basics and beyond. J. Chemom. 17
tion, Methodology, Formal analysis, Writing - original draft. Jesus (8–9), 480–502.
Flores-Cerrillo: Writing - review & editing, Supervision.
 A. Kumar, A. Bhattacharya and J. Flores-Cerrillo / Computers and Chemical Engineering 136 (2020) 106756
Qin, S.J., 2012. Survey on data-driven industrial process monitoring and diagnosis.
Annu. Rev. Control 36 (2), 220–234.
Reis, M., Gins, G., 2017. Industrial process monitoring in the big data/industry 4.0
era: from detection, to diagnosis, to prognosis. Processes 5 (3), 35. doi:10.3390/
pr5030035.
Ricker, N.L., 1988. The use of biased least-squares estimators for parameters in dis-
crete-time pulse-response models. Ind. Eng. Chem. Res. 27 (2), 343–350.
Severson, K., Chaiwatanodom, P., Braatz, R.D., 2016. Perspectives on process moni-
toring of industrial systems. Annu. Rev. Control 42, 190–200.
Simon, D.L., Litt, J.S., 2011. A data filter for identifying steady-state operating points
in engine flight data for condition monitoring applications. J. Eng. Gas Turbine
Power 133 (7), 71603.
Tsung, F., 20 0 0. Statistical monitoring and diagnosis of automatic controlled pro-
cesses using dynamic PCA. Int. J. Prod. Res. 38 (3), 625–637.
Venkatasubramanian, V., Rengaswamy, R., Kavuri, S.N., 2003. A review of process
fault detection and diagnosis: part ii: qualitative models and search strategies.
Comput. Chem. Eng. 27 (3), 313–326.
19
Venkatasubramanian, V., Rengaswamy, R., Kavuri, S.N., Yin, K., 2003. A review of
process fault detection and diagnosis: part iii: process history based methods.
Comput. Chem. Eng. 27 (3), 327–346.
Venkatasubramanian, V., Rengaswamy, R., Yin, K., Kavuri, S.N., 2003. A review of
process fault detection and diagnosis: part i: quantitative model-based methods.
Comput. Chem. Eng. 27 (3), 293–311.
Yamamoto, T., Shimameguri, A., Ogawa, M., Hashimoto, I., Kano, M., 2004. Appli-
cation of statistical process monitoring with external analysis to an industrial
monomer plant. IFAC Proc. Vol. 37 (1), 379–384.
Yin, S., Ding, S.X., Xie, X., Luo, H., 2014. A review on basic data-driven approaches
for industrial process monitoring. IEEE Trans. Ind. Electron. 61 (11), 1–10.
Yin, S., Gao, X., Karimi, H.R., Zhu, X., 2014. Study on support vector machine-based
fault detection in tennessee eastman process. Abstract and Applied Analysis, vol.
2014. Hindawi.