Journal of Loss Prevention in the Process Industries 72 (2021) 104578

Contents lists available at ScienceDirect

Journal of Loss Prevention in the Process Industries

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

Alarm management optimization in chemical installations based on

adapted HAZOP reports
Yifei Meng a, *, Xiaomiao Song a, Dongfeng Zhao a, Qinglong Liu b
a
College of Chemical Engineering, China University of Petroleum (East China), China
b
Qingdao Oasis Environmental & Safety Technology Co., Ltd, China

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

Keywords: Most current alarm systems used in chemical installations show poor performance due to alarm flooding. This
Alarm management optimization study focuses on alarm management systems optimization using the deviation propagation relationship hidden in
Alarm flooding the hazard and operability study (HAZOP) report, which can be transformed into a critical information source for
HAZOP
alarm optimization management. More concretely, this means matching the alarm tag number with the process
Deviation propagation
deviations in the deviation column, possible cause column, and consequence column. Furthermore, a back­
tracking method and a reasoning method were established to identify the initial alarm and associated alarms.
Besides, a root fault diagnosis was carried out. A method of detecting hardware faults and unreasonable alarm
thresholds is established using alarm causality corresponding to the deviation causality and associated alarm
generation-skipping tracing method. According to the severity of the consequence corresponding to the devia­
tion, a determined alarm priority method is constructed. The results show that the deviation propagation rela­
tionship in the HAZOP report is clear, and the topological relationship is easy to build based on the deviation
propagation relationship. With comprehensive and in-depth HAZOP analysis reports in China, the alarm man­
agement optimization technology based on adapted HAZOP reports shows good prospects for application and
promotion.

1. Introduction To solve alarm flooding problems, it is necessary to consider the in­

dustrial process’s correlation characteristics, develop the corresponding
As one independent protection layer for process safety, the alarm topology model, determine the alarm’s source, and choose the correct
system plays an essential role in modern chemical installations. How­ fault path. For such instances, the operation can be switched to normal
ever, due to poor alarm management, most industrial alarm systems mode. Many scholars have researched processing topology modeling,
currently in use have poor performance, and the most common and and alarm root cause analysis (Goel et al., 2017). Topology modeling
intractable problem is alarm flooding (Niyazmand et al., 2019; Lucke methods are divided into four categories: mechanism modeling-based
et al., 2018), which may result in severe and fatal accidents. For methods, data-driven methods, knowledge-based methods, and infor­
example, in 1979, at Three Mile Island, the worst nuclear leakage ac­ mation fusion methods. Mechanism modeling-based methods use system
cident occurred due to an external disturbance of the alarm system with parameter identification, state estimation, and other methods to estab­
at least one alarm activated on average every 2~3s. Confused by lish an accurate mechanism model. The main methods include adaptive,
continuous and constant alarm signals, the operator failed to identify the multi-objective optimization, and fuzzy observer etc. Data-driven
valve abnormality, which eventually led to the accident (Skillman et al., methods establish the statistical law or distribution law of data based
2020). Wang (Wang et al., 2016) summarized five main reasons for on the production process data, various data analysis methods, and the
alarm flooding: a) complex characteristics of process correlations with a steady-state ergodic hypothesis. The main methods are as follows:
wide range of abnormal propagation; b) the random alarm threshold wavelet transform, principal component analysis (Gu et al., 2005), ca­
design, unbalanced false alarm rates, and missing alarm rates; c) the nonical variable analysis, partial least squares, neural network, and
fuzzy alarm priority and improper processing sequence; d) alarm pri­ exploratory data analysis（; Bezerra et al. (2019); Goel et al. (2019) etc.
ority; and e) absence of alarm performance evaluation. Knowledge-based methods mainly rely on the existing process

* Corresponding author.
E-mail address: myf213@163.com (Y. Meng).

https://doi.org/10.1016/j.jlp.2021.104578
Received 1 January 2021; Received in revised form 1 June 2021; Accepted 9 June 2021
Available online 21 June 2021
0950-4230/© 2021 Elsevier Ltd. All rights reserved.
Y. Meng et al. Journal of Loss Prevention in the Process Industries 72 (2021) 104578

knowledge and expert experience to establish the industrial process 2. Establishment of process topology based on HAZOP analysis
association model or causal model (Yang et al., 2012). Knowledge-based report
methods mainly include expert systems, causality diagrams, signed di­
graphs, fault trees, and other methods. Information fusion methods 2.1. Theoretical basis of the establishment of the topological relationship
focus on using existing knowledge and data to improve the alarm
model’s accuracy (Gao et al., 2015; Chiang et al., 2003). Table 1 sum­ In HAZOP analysis, potential process deviations are first identified
marizes the advantages and disadvantages of the four methods and by matching process parameters with leading words. Then, technical
related research progress. professionals analyze the causes, consequences, existing risk control
Typical knowledge-based models include SDG model (Wang et al., measures, and the process of proposed control measures for every de­
2013), multistage flow model (Us et al., 2011), causality diagram viation through collaborative brainstorming. Deviation causes typically
network (Hu et al., 2015), Bayesian network (Mori et al., 2014), and peri include process deviations, equipment and instrument failures, and
network model (Kezunovic et al., 2009). As stated in Table 1, management control failure. Consequences are usually expressed in the
knowledge-based methods can deeply mine the process information of form of process deviations, accident scenarios, and adverse conse­
internal structure, process mechanism, and the correlation between quences. The propagation relationship between the process parameters
variables. However, establishing topological models of industrial pro­ comprises the cause and consequence’s process deviation and the cur­
cesses often relies on plant connectivity information and expert expe­ rent deviation, and if these process parameters are configured with a
rience. For processes without sufficient knowledge and experience, this threshold alarm in the system, there is a correlation between these
method is still helpless. In addition, the heavy workload of the estab­ alarms. Furthermore, the correlation can be used to analyze the alarm
lishment of topological models limits its application in large-scale in­ signal and optimize the alarm system. In the process of the cause or
dustrial processes (Swann et al., 1995). HAZOP analysis is a structured consequence analysis, some causes or consequences may be expressed in
analysis method for hazards and operability based on process parameter the form of fault or accident scenarios, but they may also be related to
deviation. It is one of the most widely used evaluation methods that can process deviation (alarm). In the correlation process, the situation
investigate every node comprehensively and systematically. In recent should be considered to identify the potential process deviation
years, China has issued a series of guidance documents to promote associations.
HAZOP analysis in hazardous chemicals (WSCO 26 (2008); SAWS
dpt.3124 (2009); SAWS dpt.3191 (2011); SAWS dpt.3186 (2010); SAWS 2.2. Transformation process of the HAZOP analysis report
dpt.3 93 (2011)). After several years of implementation, HAZOP anal­
ysis has been implemented across China. All petrochemical plants in Section 2.1 explains the HAZOP analysis report that contains the
large state-owned enterprises (SOE), such as PetroChina and Sinopec, process of topological relationship. In practical applications, standard­
continuously carry out HAZOP analysis. The coverage rate of HAZOP ized data must be processed in the HAZOP analysis, enabling computers
analysis among non-SOEs is also extremely high. Because the HAZOP to import and analyze the data. This process can be called the deviation
analysis process includes the systematic analysis of process flow, this alarm signal mapping process (Fig. 1).
paper depends on the HAZOP analysis report to quickly build a process Firstly, according to the selected research object, the latest version of
topology model and provides progressive research on its applications for the HAZOP analysis report is obtained. Then, node by node deviation is
alarm system optimization in industrial applications. analyzed and processed. The full process contains three smaller pro­
cessing steps. The first step is to confirm whether the recommended
measures corresponding to the current deviation have been imple­
Table 1
mented. It should be noted that some alarm settings may be included in
Method comparison.
the proposed measures. If the proposed measures have been imple­
Methods Advantages Disadvantages mented, the presence of the alarm should be considered. The second step
Mechanism Based on the reaction Due to the continuous change is to determine the alarm signal corresponding to the deviation, cause,
modeling- mechanism, and the process and fluctuation of the and consequence combined with the existing protection measures, the
based methods parameters can be accurately production process and the
recommended measures implemented, and the process control situation.
simulated external random disturbance,
it is not easy to establish an In this step, it should be noted that some deviation causes and conse­
accurate model, limiting the quences in the original report are not expressed in the form of deviation,
method’s practical but the corresponding alarm signals must be identified. The third step is
application. to add one column respectively following the process deviation column,
Data-driven Effectively use the existing It is difficult to find a
cause column, and consequence column to fill in the corresponding
methods process data and make the reasonable statistical law
results more reliable. because of the complex alarm tag number of front column deviation. The HAZOP analysis table
structure and uneven quality can be obtained by implementing the above process, as shown in
of the field data. Even after Table 2.
preprocessing, the data may
still have complex
characteristics. 2.3. Research on alarm optimization based on improved HAZOP analysis
Knowledge-based Knowledge-based methods For processes without
methods can deeply mine the process sufficient knowledge and 2.3.1. Ideas of alarm optimization
information of internal experience, Knowledge-based After processing the HAZOP analysis, the correlation relationship
structure, process methods are helpless.
mechanism, and the Moreover, these qualitative
between the relevant process deviations (part of the corresponding
correlation between models do not contain any alarm signals) can be obtained. After processing and analyzing the alarm
variables. quantitative information and data, the optimal management of the alarm signal, including the asso­
cannot identify essential ciated alarm identification and fault diagnosis, false alarm identification
factors.
and threshold optimization, alarm priority determination, etc., can be
Information The combination of these Information fusion methods
fusion methods methods can overcome the lack the research on the realized. Fig. 2 shows the conceptual framework of alarm optimization.
difficulties of traditional method of fault diagnosis and
fault diagnosis methods fault path identification for 2.3.2. Associated alarm identification and fault diagnosis
complex industrial processes Reasons for alarm flooding are the complex process correlation
under the guidance of quality.
characteristics and a wide range of abnormal propagation. e.g., when a

2
Y. Meng et al. Journal of Loss Prevention in the Process Industries 72 (2021) 104578

analysis deviation (deviation in Table 2) is retrieved from the improved

HAZOP analysis table according to the tag number of alarm signal i, then
all the cause items in the deviation analysis line are extracted. Next, the
reason items are divided into three categories: the process deviation
corresponding to the alarm tag number, the process chip deviation
referring to the false alarm tag number, and the third category of non-
process deviation. If there is one or more alarm signals j correspond­
ing to the first type of reason item in the current alarm signal set, the
current alarm i is considered as the associated alarm of the alarm j. If all
alarms corresponding to the first type of reason item do not appear, the
analysis line of each reason item is retrieved from the improved HAZOP
report, and the cause item of the reason item is circularly divided. The
purpose of this cycle is to trace the alarm signal of the next generation
forward. In Fig. 5, m<2 indicates that the maximum intergenerational
traceability depth is two levels, and the number of layers can be defined
in practical applications. After the first type of cause item analysis is
completed, the analysis line of the reason item in the second type is
retrieved from the improved HAZOP report, and the next level analysis is
cycled. If the associated alarm has not been found, the alarm is
considered as the initial alarm and is included in the initial alarm set.
Similarly, if all cause items corresponding to process alarm i do not
belong to the first or second category, then alarm i is directly identified
as the initial alarm and is included in the initial alarm set.
The advantage of that backtracking method is that it only analyzes
the current alarm signal set and requires a small amount of data pro­
cessing. However, the analysis requires a batch cycle calculation, which
takes a long time to process. The rationale of the reasoning method is
contrary to the backtracking method. By analyzing all current alarms
and retrieving the consequence items, a potential predictive alarm set is
formed. If alarm j appears in the predicted alarm set in the following
stage, it is determined as an associated alarm; otherwise, it is defined as
an initial alarm. The process must analyze the alarm data at the previous
moment, so the reasoning method enjoys the advantages of large data
processing capacity and simple calculations. This specific analysis pro­
cess is shown in Fig. 4.
Once it is defined as the initial alarm, it must be processed quickly.
The process topology relationship constructed by this method also
provides a direct screening range for the causes of relevant alarms.
Fig. 1. Adapting process of the HAZOP report.
2.3.3. False alarm identification and alarm threshold optimization
low-pressure deviation occurred in the EDC reactor, due to the deviation There are a large number of false alarms in the alarm signal due to
transmission through the system, a series of other deviations, such as hardware failure and chattering alarms caused by unreasonable
low EDC flow, high temperature of the VCM cracking furnace, and high threshold designs. At present, alarm flooding is typically solved by
pressure of the VCM cracking furnace tube, maybe caused over a short threshold optimization, hardware maintenance, and delay alarm tech­
period. Then the corresponding alarm signals (if any) may also be suc­ nology. In this paper, the process topology model that is established by
cessively triggered over a short time. In situations like this, if the staff improved HAZOP analysis, can also detect false alarms for unreasonable
cannot find the root cause quickly and execute a proper response, it may threshold design and instrument failure, as well as further guide
lead to further deterioration of the situation, such as furnace tube threshold optimization and maintenance guidance.
rupture/leakage in this case. According to the constructed process to­ The identification of false positives can be implemented by
pology, two methods can be obtained to identify association alarms, combining the methods of backtracking and reasoning. If an alarm
namely, the backtracking and reasoning methods. signal is found to be the initial alarm through the backtracking method,
The basic idea of the backtracking method is to use current real-time but its consequence deviations that judged by reasoning method never
alarm data to analyze the relationship between alarm signals. Fig. 3 appear (the corresponding alarm signals are not triggered) in a period of
shows the analytical process of the backtracking method. There are a time. If the time period is selected reasonably, it is clear that the alarm
few steps to analyze any alarm signal i. Firstly, the corresponding signal is likely to be a false positive. The alarm tag number can be

Table 2
Adapted HAZOP report.
Serial Deviation Alarm tag Cause Alarm tag Consequence Alarm tag
number number number number

5.4 EDC flow is FI-******* DC reactor pressure is low PI-******* Temperature of the VCM cracking TI-******
low furnace is high
EDC sampling connection (upstream) is EDC leaking into the environment
not closed
Contamination of EDC cooler

*The complete table also contains columns of existing protection measures and recommended measures, which are unnecessary in the data processing process.

3
Y. Meng et al. Journal of Loss Prevention in the Process Industries 72 (2021) 104578

Fig. 2. The idea of alarm optimization based on an adapted HAZOP report.

Fig. 3. Associated alarm identification process by the backtracking method.

4
Y. Meng et al. Journal of Loss Prevention in the Process Industries 72 (2021) 104578

Fig. 4. Associated alarm identification process by the reasoning methodology.

Table 3
Adapted HAZOP table with risk analysis.
Serial number Header Key parameter

1 Serial number
2 Deviation √
Fig. 5. Associated alarm intergenerational tracing process. 3 Deviation alarm tag number √
4 Reason √
identified and checked in the subsequent maintenance plan. 5 Cause alarm tag number √
6 Consequence
In the associated alarm recognition stage, this paper derives a
√
7 Consequence alarm tag number √
generation-skipping tracing mechanism. As shown in Fig. 5, alarm sig­ 8 Existing protection measures
nals corresponding to deviation a, c and d but not deviation b are trig­ 9 Possibility level
gered, according to the principle of generation-skipping traceability, 10 Severity level √
11 Risk level
there are specific reasons to believe that the fault path is a →b → c → d,
12 Suggested protective measures
and the alarm corresponding to b deviation fails to trigger due to
hardware damage or unreasonable threshold setting. Then an alarm tag
number of b deviation can be identified and verified in the subsequent 3. Case studies
maintenance plan.
3.1. Case introduction
2.3.4. Alarm priority determination
After the identification of associated alarms, the current initial alarm In this paper, a HAZOP case of a pilot plant of vinyl chloride (VCM)
set can be obtained. All alarms in the initial alarm set are the root causes production process from CCPS Guidelines for Hazard Evaluation Pro­
of other alarms. Therefore, such alarms should be treated as important cedures (2008) was selected to demonstrate the process. The analysis
alarms. Moreover, the integration of HAZOP, LOPA, and even SIL study node is the cracking furnace unit. Table 4 is an excerpt from the original
is becoming mainstream, and many HAZOP analysis tables have been HAZOP report.
expanded compared with the traditional analysis forms, such as adding
columns for possibility, consequence severity, risk, SIL for SIFs. This
information expansion is benefit to determine alarm priority for 3.2. Process topology analysis of HAZOP report
different deviation. The danger of the current deviation can be expressed
by the value of the consequence severity level column corresponding to The topological process relationship of Table 4 is illustrated in Fig. 6,
the current deviation, and then the corresponding alarm priority can be where solid-line circles in the figure represent the process deviation, the
determined. It should be noted that the priority of the alarms in the upstream of the input arrow is the cause factor of the deviation, and the
initial alarm set is not only determined by the calculated value of the downstream of the output arrow is the consequence factor of the devi­
consequence severity level column corresponding to the alarm but also ation. For example, there are two reasons for High temperature -VCM
based on the highest value of the consequence severity level column of furnace (5.7) of VCM cracking furnace, one is High flow-nature gas (5.2),
the associated alarm. As shown in Fig. 5, the alarm priority corre­ and the other is Low flow -EDC (5.4). There are two consequences: High
sponding to deviation a shall be determined by the highest consequence pressure-VCM furnace tube (5.9) and tube leakage/rupture (5.12). The
severity level value of deviations a, b, c, and d. Table 3 adds the alarm tag dotted circle in the figure represents the special process deviations. The
number to the header of the HAZOP analysis table commonly used in special process deviation cannot be expressed by the terms high or low,
engineering practice, which constitutes the modified HAZOP report and they have no corresponding alarm signal. However, such deviation
header. The parameter √ is the critical application parameter of this may lead to further process deviations, such as furnace tube leakage/
method. rupture (5.12). In the process control system, there is no special
With the improved HAZOP report, if there is no alarm set for the measurable parameter to express the deviation. However, after the de­
deviation with high consequence severity level or risk level and no alarm viation occurs, it may cause an excessive concentration of EDC in the
set for the corresponding deviation of cause column and consequence environment and give a concentration alarm. The regular octagon in the
column, then it can be assumed that the alarm setting is unreasonable to figure indicates the factors caused by non-upstream process deviation,
a certain extent. such as equipment or instrument failure, management, and control
failure. From the figure, the hierarchical relationship of the process

5
Y. Meng et al. Journal of Loss Prevention in the Process Industries 72 (2021) 104578

Table 4
HAZOP report example (Extract).
Item Deviation Causes Consequences Safeguards Actions

5 Cracking furnace - VCM furnace (normal operation; raise EDC to 900 C , 160 psig to make VCM, flow 1200 lb/h)
◦

5.1 High flow-EDC High Pressure in Low conversion of EDC to VCM. High Determine if incinerator has the capacity
DC reactor (3.5) carry-over of EDC to incinerator with to destroy EDC when EDCNCM
potential EDC release to the conversion is low. Consider installing an
environmen EDC stack monitor.
Low process temperature in the VCM
furnace (Item 5.8)
5.2 High flow- Gas FCV fails open, Possible loss of flame and potential TIC controls gas supply Consider a PAH and high pressure
nature gas natural gas supply explosion shutdown for natural gas using a positive
pressure high isolation valve；Consider installing an
TIC fails low signal High process temperature in the VCM independent TAHH and high furnace
furnace (Item 5.7) discharge temperature shutdown ；
Potential tube damage and possible Consider installing a flame scanner and
fire if tube ruptures. Excess by- loss of flame shutdown
products in product stream (Item 5.12)
5.4 Low flow -EDC low pressure in DC High process temperature in the VCM Operator present when EDC sample Consider installing a PAL on the direct
reactor (Item 3.6) furnance (Item 5.7) connection used chlorination reactor
EDC sample High production of by-products during
connection EDC cracking. Potential furnace tube
(upstream) left damage and possible fire if tube
open ruptures. EDC release to the
environment (Item 5.12)
fouling of the EDC
cooler
5.7 High High natural gas High pressure in the VCM furnace Furnace tube skin TAH Consider installing an independent
temperature flow (Item 5.2) tubes (Item 5.9) Product discharge TAH and TAHH with TAHH and high furnace discharge
-VCM furnace hardhired shutdown of plant temperature shutdown
Low EDC flow High production of by-products during Furnace tubes designed to withstand very
(Item 5.4) EDC cracking. Potential furnace tube high temperatures
damage and possible fire if tube
ruptures. Furnace damage (Item 5.12)
5.9 High pressure- High pressure in Potential furnace tube damage and
VCM furnace the VCM furnace possible fire if tube ruptures. EDC
tube tubes (Item 5.7) release to the environment (Item 5.12)
5.12 Tube leakage/ Fouling EDC leaking into the environment (6. Fire monitors in furnace area; Emergency Verify adequate quality assurance
rupture *) response team trained for fire fighting; program for all furnace tubes
Corrosion Furnace fire with subsequent release Ethylene and chlorine supplies can be
of EDC to the environment shut off remotely; Tubes inspected and
Bad weld Potential major equipment damage welds x-rayed prior to service; Furnace
High flow - natural will only be run for short periods (few
gas (Item 5.2) days) over the next year. Fouling should
LOW flow - EDC be minor; Tube material of construction
(Item 5.4) adequate for EDC, chlorine, and ethylene
High temperature service
(Item 5.7))

deviation transfer process is not strictly required in the HAZOP analysis should be searched and confirmed whether the reason items with alarm
process, so there may be a cross-level phenomenon of cause factors. In tag number in the cause column have any alarm response. In this case,
this example, in addition to High temperature -VCM furnace (5.7), fouling, due to boundary problems, it was assumed that all reason items with
corrosion and bad weld may also cause tube leakage/rupture (5.12). At the alarm tag numbers in the Low pressure-DC reactor analysis line were not
same time, two reasons for High temperature -VCM furnace (5.7) are High triggered. Therefore, the alarm signal corresponding to low flow - EDC
flow-nature gas (5.2) and Low flow -EDC (5.4). Although this cross-level can be recognized as an initial alarm. Then, it was found that the alarm
phenomenon may seem confusing to the system, it has no essential corresponding to the high temperature - VCM furnace in the cause item
impact on the follow-up analysis. It should be noted that Fig. 6 is only was not triggered through the analysis of the high pressure-VCM furnace,
used to express the topological process relationship contained in the so it was suspected that it was the initial alarm. However, the alarm
HAZOP analysis report, and it is exclusively utilized for the principle corresponding to low flow- EDC at the next higher level was triggered,
analysis of this paper. The actual analysis process may differ. which indicates that high pressure-VCM furnace is the probable associated
alarm for low flow - EDC. After further analysis of the EDC concentration
3.3. Alarm optimization analysis high -external environment, there was no corresponding alarm tag number
for the furnace tube leakage/rupture in the cause item. Furthermore, high
To demontrate the optimimization analysis process, two alarm signal pressure-VCM furnace was triggered, which indicated that EDC concen­
status sets at different points in time were imaged as shown in Table 5. tration high -external environment was the associated alarm of high pres­
For point 1, the backtracking method was used to analyze all trig­ sure- VCM furnace, and high pressure- VCM furnace is the associated alarm
gered alarms. Low flow- EDC alarm was analyzed first, and the analysis of low flow-EDC. Consequently, among three alarm signals, low flow- EDC
line with deviation was found through the retrieval of the adapted is the initial alarm, and the other two alarms are its associated alarms.
HAZOP report. It was found that Low pressure-DC reactor in the corre­ The HAZOP analysis report provided in this case does not include the
sponding three causes related to an alarm tag number, but the alarm was consequence level column, so the priority determination based on the
not triggered. According to the principle of generation-skipping trace­ consequence level was not considered. According to the alarm priority
ability, the analysis line of Low pressure-DC reactor in the HAZOP report principle, low flow-EDC as the initial alarm should be given priority,

6
Y. Meng et al. Journal of Loss Prevention in the Process Industries 72 (2021) 104578

Fig. 6. Process topology diagram for the HAZOP report, as shown in Table 4.

4. Conclusion and future directions

Table 5
Alarm signal status at two different points in time.
In this paper, the causal relationship between the cause-deviation-
Serial Deviation Alarm signal Alarm signal result columns of the HAZOP analysis is used to research alarm opti­
number status at point 1 status at point 2
mization management in chemical enterprises. The conclusions are as
1 Pressurelow - DC reactor no no follows:
2 Low flow - EDC yes no
3 High flow -Fuel gas no yes
4 Temperature high - VCM no no
(1) The deviation transfer relationship in HAZOP analysis report is
furnace clear, and it is relatively simple to construct process topology
5 Pressure high - VCM furnace yes no based on this analysis;
6 EDC concentration high yes no (2) HAZOP analysis has been comprehensively promoted. The alarm
-external environment
optimization management technology based on HAZOP has a
good foundation for promotion and implementation on a large
while the other two alarms as associated alarms have relatively low scale;
priority, but this conclusion is not absolute. In most cases, the confir­ (3) The technology of alarm optimization management based on
mation of priority should still focus on the consequence severity. improved HAZOP analysis has apparent practicality in related
During the analytical process of the above backtracking method, if alarm identification, root fault diagnosis, false alarm identifica­
the alarm corresponding to low flow- EDC is determined as the initial tion, alarm threshold optimization, and alarm priority
alarm by the backtracking method, the potential cause according to the determination.
topological relationship is more likely to be EDC sample connection
(upstream) left open or fouling of the EDC cooler. According to the prompt, In the research process, the following problems were not intention­
the staff can focus on the investigation and confirmation of these two ally considered. Therefore, further research will be in the follow-up
causes, which can greatly improve the efficiency of the root cause fault study:
diagnosis.
When performing the alarm state associated alarm analysis shown in (1) Currently, this technology is only suitable for installations with
the case above, it was found that the alarm signal of high temperature- proven high-quality HAZOP reports. For installations without
VCM furnace was not triggered when performing associated alarm high-quality HAZOP reports, a quality review of the HAZOP
analysis. However, high temperature-VCM furnace likely caused sensor report should be done before applying this technology. In future
damage or an unreasonable alarm threshold. Therefore, it is necessary to work, the HAZOP report quality quick review procedure should
verify and confirm the alarm. be developed as the pre-module of this technology.
If the high flow-fuel gas in point 2 triggers a continuous, stable alarm (2) This paper assumes that the system alarm threshold setting is in
for a relatively long period of time, the subsequent and intergenerational an appropriate state in the two parts of associated alarm recog­
high temperature-VCM furnace, high pressure-VCM furnace, and EDC con­ nition: hardware fault identification and alarm threshold opti­
centration high -external environment do not alarm, and it is reasonable to mization. Although the intergenerational traceability mechanism
suspect that the alarm was a false alarm. was developed to eliminate local irrationality, if the overall alarm

7
Y. Meng et al. Journal of Loss Prevention in the Process Industries 72 (2021) 104578

setting is out of order, the technology cannot be used or be Chiang, L.H., Braatz, R.D., 2003. Process monitoring using causal map and multivariate
statistics: fault detection and identification. Chemometr. Intell. Lab. Syst. 65 (2),
applied to the system debugging stage.
159–178. https://doi.org/10.1016/S0169-7439(02)00140-5.
(3) Further research can work on the efficient use of massive alarm Gao, Z., Cecati, C., Ding, S.X., 2015. A survey of fault diagnosis and fault-tolerant
data self-learning to optimize the alarm threshold. techniques—part i: fault diagnosis with model-based and signal-based approaches.
(4) This work focuses on theoretical analysis and process develop­ IEEE Trans. Ind. Electron. 62 (6), 3757–3767. https://doi.org/10.1109/
TIE.2015.2417501.
ment. A software/platform will be developed using Visual Studio Goel, P., Datta, A., Mannan, M.S., 2017. Industrial alarm systems: challenges and
or some other development environment in future work. The opportunities. J. Loss Prev. Process. Ind. 50, 23–36. https://doi.org/10.1016/j.
software can be used by DCS operators as well as safety engineers. jlp.2017.09.001.
Goel, P., Pistikopoulos, E.N., Mannan, M.S., Datta, A., 2019. A data-driven alarm and
The system is expected to have a high demand for industrial ap­ event management framework. J. Loss Prev. Process. Ind. 62, 103959. https://doi.
plications in view of the current chaotic situation of enterprise org/10.1016/j.jlp.2019.103959.
alarm management. In terms of business costs, the investment of Gu, X.B., Zhu, Q.X., 2005. Alarm system reconstruction algorithm and application for
process industry based on matter-element. Control Decis. 20 (9), 1006–1011.
the enterprise includes one-time labor costs for reviewing and https://doi.org/10.13195/j.cd.2005.09.48.guxb.009.
adapting existing HAZOP reports (far less than the cost of HAZOP Hu, W., Wang, J., Chen, T., 2015. A new method to detect and quantify correlated alarms
analysis), software purchase costs, computer and network hard­ with occurrence delays. Comput. Chem. Eng. 80, 189–198. https://doi.org/10.1016/
j.compchemeng.2015.05.028.
ware and maintenance costs needed to support software opera­ Kezunovic, M., Guan, Y., 2009. Intelligent alarm processing: from data intensive to
tion, and small labor costs caused by changes in HAZOP form information rich. Hawaii International Conference on System Sciences. IEEE.
information caused by system changes. https://doi.org/10.1109/HICSS.2009.268.
Lucke, M., Chioua, M., Grimholt, C., , et al.Thornhill, Nina F., 2018. On improving fault
detection and diagnosis using alarm-range normalization. IFAC-PapersOnLine 51,
Author contribution statement 1227–1232. https://doi.org/10.1016/j.ifacol.2018.09.695.
Mori, J., Mahalec, V., Yu, J., 2014. Identification of probabilistic graphical network
Yifei Meng: Study concepts, Study design, Manuscript preparation. model for root-cause diagnosis in industrial processes. Comput. Chem. Eng. 71,
171–209. https://doi.org/10.1016/j.compchemeng.2014.07.022.
Song Xiaomiao: Literature research, Manuscript editing. Zhao Dongfeng: Niyazmand, T., Izadi, I., 2019. Pattern mining in alarm flood sequences using a modified
Study concepts, Manuscript revision/review. Liu Qinglong: Study PrefixSpan algorithm. ISA (Instrum. Soc. Am.) Trans. 90, 287–293. https://doi.org/
design, Manuscript revision/review 10.1016/j.isatra.2018.12.050.
Skillman, G.R., Rempe, J.L., 2020. The three-mile island unit 2 accident. Reference
Module in Earth Systems and Environmental Sciences. https://doi.org/10.1016/
Declaration of competing interest B978-0-12-409548-9.12146-3.
State Administration of Work Safety dep.3, 2009. Guidance of State Administration of
Work Safety on Further Strengthening the Work Safety Standardization of Hazardous
The authors declare that they have no known competing financial Chemicals Enterprises. No. 124. https://www.mem.gov.cn/gk/gwgg/agwzlfl/yj_0
interests or personal relationships that could have appeared to influence 1/200906/t20090625_242155.shtml.
the work reported in this paper. State Administration of Work Safety dpt. 3 & Ministry of Industry and Information
Technology, 2010. Implementation Opinions of Ministry of Industry and Information
Technology & State Administration of Safety on Implementing the Circular of the
Acknowledgment State Council on Further Strengthening the Work Safety of Enterprises for Hazardous
Chemicals Enterprises. No.186. https://www.mem.gov.cn/gk/gwgg/agwzlfl/gfxwj/
2010/201011/t20101110_243088.shtml.
The work is supported by the National Natural Science Foundation of
State Administration of Work Safety dpt.3, 2011a. Notice of the State Administration of
China (21406271) and the key research and development program of Work Safety on Printing and Distributing the Standardization Evaluation Standards
Shandong Province(2019JZZY020502). The authors thank TopEdit for the Work Safety of Hazardous Chemicals Working Units. No. 93. https://www.
(www.topeditsci.com) for its linguistic assistance during the preparation mem.gov.cn/gk/gwgg/agwzlfl/gfxwj/2011/201106/t20110622_243041.shtml.
Swann, C.D., Preston, M.L., 1995. Twenty-five years of HAZOPs. J. Loss Prev. Process.
of this manuscript. Ind. 8 (6), 349–353. https://doi.org/10.1016/0950-4230(95)00041-0.
Us, T., et al., 2011. Fundamental principles of alarm design. International Journal of
References Nuclear Safety and Simulation 2 (1), 44–51.
Wang, J., Yang, F., Chen, T., Shah, S.L., 2016. An overview of industrial alarm systems:
main causes for alarm overloading, research status, and open problems. IEEE Trans.
State Administration of Work Safety dpt.3, 2011. Notice of the State Administration of Autom. Sci. Eng. 1045–1061. https://doi.org/10.1109/TASE.2015.2464234.
Work Safety on Issuance of the "12th Five-Year Plan" for the Safe Production of Work Safety Committee Office of the State Council, 2008. Guidance of Work Safety
Hazardous Chemicals. No. 191. https://www.mem.gov.cn/gk/gwgg/agwzlfl/tz_01/ Committee Office of the State Council on Further Strengthening the Safe Production
201112/t20111215_236000.shtml. of Hazardous Chemicals. No. 26. https://www.mem.gov.cn/awhsy_3512/awh
Bezerra, A., Silva, I., Guedes, L.A., Silva, D., Leitão, G., Saito, K., 2019. Extracting value bgswj/200810/t20081007_248159.shtml.
from industrial alarms and events: a data-driven approach based on exploratory data Yang, F., Xiao, D.Y., 2012. Progress in root cause and fault propagation analysis of large-
analysis. Sensors 19 (12), 2772. https://doi.org/10.3390/s19122772. scale industrial processes. J. Contr. Sci. Eng. 478373. https://doi.org/10.1155/
CCPS/AIChE, 2008. Guidelines for Hazard Evaluation Procedures. Willy interscience. 2012/478373.
https://doi.org/10.1002/9780470924891.