Study on Risk Analysis of a Rotary Kiln-Based Activated Carbon Manufacturing Process Using Fuzzy-FMEA

Abstract

Rotary kiln-based activated carbon production combines high-temperature operation with flammable/reducing gases, carbonaceous dust, and downstream off-gas treatment and acid/base washing, creating complex escalation pathways. This study prioritizes safety improvements by applying classical failure modes and effects analysis (FMEA) and a transparent Fuzzy-FMEA framework to 18 representative failure modes (six each for kiln/activation, acid/base handling, and atmosphere/control). Five experts evaluated Severity, Occurrence, and Detection on a 10-point scale. The fuzzy model used triangular membership functions (L/M/H), a monotonic 27-rule base, Mamdani max–min inference, and centroid defuzzification to compute a continuous fuzzy risk priority number (FRPN, 0–10). Classical FMEA identified dust explosion (RPN = 405), temperature control failure (RPN = 378), and off-gas leakage (RPN = 324) as the highest-ranked risks. Fuzzy-FMEA preserved the top-risk group while more strongly highlighting barrier-related risks, placing off-gas leakage, instrumentation/interlock failure, and electrostatic ignition control alongside dust explosion (FRPN 9.221–9.332). The rankings were strongly correlated (Spearman ρ = 0.871; Kendall τ = 0.752), yet mid-risk items were rearranged (mean |Δrank| = 2.06; max = 5), improving discrimination within tied RPN clusters. The five highest-priority scenarios were reconstructed into actionable engineering packages, including dust and ignition control, off-gas integrity linked to shutdown logic, interlock proof testing and bypass management, and independent protection layers for kiln temperature control.

1. Introduction

Activated carbon is widely used in water treatment, air pollution control, solvent recovery, catalysis, energy storage, and advanced materials [1,2,3,4]. As industrial demand increases, production facilities have continued to scale up and shift toward continuous operation [5,6]. Continuous processes using a rotary kiln offer advantages in productivity and operational stability; however, their inherent characteristics, including high-temperature heat sources, flammable and reducing gases, carbonaceous dust, and downstream off-gas treatment and cooling systems, create complex, high-hazard systems in which fires, explosions, leaks, poisoning, and asphyxiation can escalate into major accidents [5,6]. In addition, acid and alkali handling during washing and neutralization introduces chemical hazards (corrosion, leakage, and exothermic mixing) that can further amplify overall process risk [7].

Although numerous process hazard analysis methodologies (e.g., Hazard and Operability Study, What-If analysis, and Layer of Protection Analysis) have been developed and applied, the key issue, from an industrial decision-making perspective, is risk prioritization, i.e., deciding where to allocate limited resources first [8,9]. Accordingly, failure modes and effects analysis (FMEA) is practical because it systematically identifies failure modes in processes and equipment, links causes, effects, and existing controls, and supports risk prioritization using Severity (S), Occurrence (O), and Detection (D) factors [10,11]. By organizing failure pathways according to process steps and clarifying where corrective actions should be implemented, FMEA functions as a proactive safety management tool for complex, high-hazard systems [10].

However, the risk priority number (RPN), widely used in classical FMEA, is susceptible to subjective expert judgment, and its multiplicative structure often produces tie-ranking issues in which different risk profiles converge to the same RPN [10,11,12]. In addition, the discrete 1–10 scoring system can create boundary discontinuities, and it is difficult to sufficiently reflect interactions among S/O/D factors or the ambiguity inherent in risk perception [10,11,12]. In particular, scenarios with poor detectability (high D) often warrant higher safety investment in practice, but this managerial implication may not be fully revealed by a simple multiplicative index [11,12].

Fuzzy set theory can mathematically handle ambiguity and uncertainty by representing linguistic judgments (e.g., low, medium, and high) as membership degrees [13]. The approach of fuzzifying input variables through membership functions, estimating output risk via if–then rule-based fuzzy inference, and then obtaining a single scalar through defuzzification is suitable for process risk assessments where qualitative and quantitative information coexist [13,14,15]. In particular, the combination of triangular membership functions (TFNs), Mamdani inference, and centroid defuzzification is widely used in industrial applications of Fuzzy-FMEA due to its simple structure and high interpretability [14,15,16].

Fuzzy-FMEA, which combines FMEA with fuzzy logic, has been developed to address tie ranking and uncertainty compression issues inherent in classical FMEA while enhancing the resolution of risk prioritization [16,17]. However, applications of a transparent Fuzzy-FMEA configuration to rotary kiln-based activated carbon manufacturing remain limited. This gap is important because barrier-dependent failure modes in this process (e.g., off-gas integrity, instrumentation/interlock reliability, and ignition source control) can share similar classical RPN values while implying different escalation pathways and management priorities. Therefore, a method that reduces tie compression while preserving interpretability is needed for this process. Previous studies have shown that reproducibility and interpretability can be strengthened through explicit membership function diagrams, rule bases, and comparison graphs/tables between RPN and FRPN [17,18], while advanced evaluation frameworks in adjacent resource industries have likewise emphasized that analytical outputs should support practical deployment and operational decision-making [19].

This study examines risk prioritization in a rotary kiln-based activated carbon manufacturing process [20]. Process hazards are structured across three domains: (i) rotary kiln and activation operation, (ii) acid/base handling, and (iii) process atmosphere and control, including temperature, overpressure, gas composition, and oxygen concentration [21]. Classical FMEA and Fuzzy-FMEA are applied in parallel to compare risk rankings and to identify priority improvement measures based on the highest FRPN items [10,16]. Importantly, this study does not stop at quantifying risk priorities; rather, it highlights that the top risks in the rotary kiln-based activated carbon process tend to escalate into accidents through combined conditions such as dust, flammable gas, high temperature, oxygen ingress, weak sealing, and instrumentation/interlock failures [22,23,24,25,26]. These high-risk items are reconstructed into representative accident scenarios, and, for each scenario, transfer pathways are interpreted from both process and equipment perspectives, including fuel sources, ignition mechanisms, oxygen supply, escalation routes, and protective layers [9]. Risk reduction strategies are then formulated as practical engineering alternatives, encompassing design improvements, off-gas system integrity, instrumentation and interlock reliability, explosion and fire protection, and degradation management. The primary contribution of this study is to translate risk prioritization outcomes into actionable guidance for process safety investment and implementation.

2. Materials and Methods

2.1. Process Scope and Definition of Failure Modes

The present study focuses on a rotary kiln-based activated carbon manufacturing process. The process scope includes: (1) raw material feeding and pretreatment; (2) carbonization and activation in the rotary kiln; (3) cooling and product conveying and storage; (4) off-gas treatment, including dust collection, afterburning/oxidation, scrubbing, and discharge; (5) acid and alkali washing and neutralization; and (6) wastewater pH management and discharge or reuse. The hazard analysis focused on high-temperature reaction/transfer risks in the rotary kiln and activation section, leakage/corrosion/exothermic mixing risks associated with acid/base handling, and failures in controlling the process atmosphere, such as off-gas composition, oxygen concentration, pressure/flow, and instrumentation/interlock failures.

Failure modes were identified by decomposing process nodes based on the process flow diagram (PFD). The selection was guided by three criteria: (i) the presence of combined hazards with high escalation potential, such as dust accumulation coupled with an ignition source and oxygen ingress, or off-gas leakage combined with weak ventilation or failure of detection and shutdown barriers; (ii) the ability to evaluate Severity, Occurrence, and Detection (S/O/D) based on existing control measures; and (iii) balanced representation across categories, with six items each for rotary kiln and activation, acid/base handling, and atmosphere/control, resulting in a total of 18 items. For each failure mode, the causes, effects, and current controls were organized in a standardized format to support subsequent scoring.

2.2. Expert Panel and Evaluation Procedure

Because the S/O/D inputs in both FMEA and Fuzzy-FMEA depend on expert judgment, a single expert panel (n = 5) was established to ensure reliability, domain relevance, and a common evidence base for direct method comparison. Panel members were selected based on professional background, role, and experience. The panel included experts with at least 10 years of experience in process operation (rotary kiln operation), equipment/maintenance, safety and health management, including process safety management (PSM) and risk assessment, chemical handling, and environment (wastewater/emissions). The same panel and the same final S/O/D set were intentionally used for both classical FMEA and Fuzzy-FMEA so that ranking differences would reflect the properties of the two prioritization schemes rather than differences between assessor groups.

The evaluation procedure consisted of: (1) process briefing and alignment on terminology/scoring criteria; (2) a review of the causes, effects, and current controls for each failure mode; (3) independent scoring of S/O/D by each expert; (4) aggregation and consensus meetings for discrepant items; and (5) final confirmation of S/O/D. The median was used as the representative score; however, when the median fell on a category boundary that could affect the final risk categorization or prioritization, or when extreme values were explainable by the presence/absence of specific controls (e.g., interlocks, redundancy, proof tests, bypass management), the final score was adjusted through evidence-based consensus. The rationale for each score was recorded for traceability. Accordingly, the comparison in this study should be interpreted as a within-panel methodological comparison rather than as validation using independent assessor groups; this design improves comparability, but limits the statistical independence between the two methods.

2.3. FMEA Scoring Framework and Ranking Approach

2.3.1. Criteria and Rationale for S/O/D Scoring

Severity (S) primarily reflects potential human impact, including fatalities, serious injuries, and minor injuries. It also accounts for the extent of equipment damage, process downtime, and environmental or regulatory consequences. O not only reflects failure frequency, but also operating modes (normal/startup/shutdown), task frequency and exposure level, and analogous failure/accident experiences (or maintenance history) in similar processes [10]. Detection (D) is defined as the ability to detect abnormalities and prevent escalation before an accident occurs, reflecting instrumentation coverage, alarm quality (false alarms/missed alarms), the presence of interlocks or automatic shutdown systems—including safety instrumented systems (SIS) where applicable—inspection/calibration/proof–test intervals, and human response time [10,24,25]. In particular, D was structured such that poorer detectability yields higher scores, ensuring that detection vulnerability increases risk [11,12]. The criteria used for assigning S/O/D scores are summarized in

Table 1. Criteria for S/O/D (10-point scale).

Category	Score	Criteria (Summary)
Severity (S)	1–2	Minor (quality impact or minor operational interruption); very low likelihood of human injury
	3–4	Limited equipment damage or minor injury; short-term process impact
	5–6	Equipment damage and brief shutdown; potential for injury; possible environmental impact
	7–8	Major (potential escalation to fire/explosion; serious injury risk; extensive damage)
	9–10	Catastrophic (possible fatalities; large-scale explosion/toxic exposure; major regulatory violation)
Occurrence (O)	1–2	Very low (almost never)
	3–4	Low (rare)
	5–6	Medium (intermittent/potentially repetitive)
	7–8	High (repetitive)
	9–10	Very high (frequent/difficult to avoid)
Detection (D)	1–2	Almost certainly detected (automatic monitoring/shutdown possible)
	3–4	Relatively easy to detect
	5–6	Typical detectability (delays/misses possible)
	7–8	Difficult to detect (inefficient inspection or blind spots exist)
	9–10	Almost impossible to detect (preemptive control is difficult)

.

Scoring was explicitly linked to current controls. For example, D for temperature control failure depends on sensor redundancy, high–high temperature interlocks, logic test intervals, and bypass management, whereas O for dust explosion depends on dust accumulation frequency (housekeeping conditions), the stability of dust collector differential pressure trends, and ignition source control. Accordingly, the S/O/D values were evaluated based on actual process conditions and the presence and performance of protective barriers.

2.3.2. RPN Calculation and Auxiliary Comparison Index

The RPN in classical FMEA was calculated as follows [10,11,12]:

$$\mathrm{R}\mathrm{P}\mathrm{N}=S\times O\times D$$

(1)

To compare with FRPN (continuous 0–10), RPN was linearly normalized to a 0–10 scale as an auxiliary index:

$${\mathrm{R}\mathrm{P}\mathrm{N}}_{\mathrm{n}\mathrm{o}\mathrm{r}\mathrm{m}}=10\times {\displaystyle \frac{\mathrm{R}\mathrm{P}\mathrm{N}-{\mathrm{R}\mathrm{P}\mathrm{N}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{{\mathrm{R}\mathrm{P}\mathrm{N}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{R}\mathrm{P}\mathrm{N}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(2)

This normalization was intended to facilitate relative comparison between the two approaches, while priority decisions were primarily based on the ranks of RPN and FRPN. If RPN_max = RPN_min, this min–max normalization becomes undefined; in that degenerate case, the auxiliary normalized index would not be interpreted, and only the common rank set would be reported. This situation did not occur in the present study (RPN_min = 120, RPN_max = 405). RPN and FRPN were ranked in descending order. When identical values occurred, the corresponding failure modes were assigned the same rank. For analyses requiring a fixed number of items, ties at the cutoff were resolved using the classical RPN as a secondary criterion (higher RPN preferred) and Severity as a tertiary criterion if needed.

2.4. Fuzzy-FMEA Method

2.4.1. Rationale for Adopting TFN–Mamdani–Centroid

This study adopts a combination of triangular membership functions (TFNs), Mamdani max–min inference, and centroid defuzzification, which are widely used in process risk assessment [13,14,15,16]. TFNs model linguistic judgments (low/medium/high) intuitively with three parameters (a, b, c), facilitating expert consensus in parameter setting [13,16]. Mamdani inference provides high explainability through a rule-based structure, and centroid defuzzification reflects the overall shape of the output fuzzy set to yield a stable scalar risk index (FRPN) [14,15]. While other operator combinations, weighting schemes, or alternative defuzzification methods could be considered, this study prioritized applicability, reproducibility, and interpretability [16].

2.4.2. Membership Functions and TFN Parameters

The input variables S/O/D have a universe of discourse of 1–10 with three linguistic terms: L (Low), M (Medium), and H (High). The output risk has a universe of discourse of 0–10 with five linguistic terms: VL (Very Low), L, M, H, and VH (Very High). The triangular membership function is defined as follows [13,16]:

$$\mu (x;a,b,c)=\left\{\begin{array}{cc}0,& x<a\\ {\displaystyle \frac{x-a}{b-a}},& a\le x\le b\\ {\displaystyle \frac{c-x}{c-b}},& b\le x\le c\\ 0,& x>c\end{array}\right.$$

(3)

TFN parameters for the input and output membership functions are provided in

Table 2. Membership (TFN) parameters for Fuzzy-FMEA.

Variable	Linguistic Term	TFN (Triangular) Membership Parameters
Input (S/O/D), domain 1–10	L	$\left(\mathrm{trimf}\left(x;1,2,4\right)\right)$
	M	$\left(\mathrm{trimf}\left(x;3,5,7.01\right)\right)$
	H	$\left(\mathrm{trimf}\left(x;6.99,9,10\right)\right)$
Output (Risk), domain 0–10	VL	$\left(\mathrm{trimf}\left(r;0,0,2\right)\right)$
	L	$\left(\mathrm{trimf}\left(r;1,3,5\right)\right)$
	M	$\left(\mathrm{trimf}\left(r;3.5,4.8,6.2\right)\right)$
	H	$\left(\mathrm{trimf}\left(r;6,7,9\right)\right)$
	VH	$\left(\mathrm{trimf}\left(r;8,10,10\right)\right)$

. Adjacent linguistic terms were designed to overlap in order to ensure continuity at boundary regions and reduce discontinuity effects. The small offsets at 6.99 and 7.01 were introduced to avoid the zero-membership dead point that can occur at the exact boundary score of 7 when adjacent triangles meet only at their endpoints in a discrete implementation. Thus, a score of 7 retains non-zero membership in both M and H, ensuring continuous rule activation; the ±0.01 shift serves as a numerical continuity device rather than an additional weighting assumption. Figure 1 illustrates both the input (S/O/D) and output (Risk) TFNs, thereby clarifying the fuzzy model applied for Mamdani inference and centroid defuzzification.

2.4.3. Rule Base Design Logic and Generation Rule

Because input linguistic terms have three levels (L/M/H), the rule base consists of 27 rules. It was designed such that the output risk increases monotonically as S/O/D increases. For this purpose, an ordinal function was defined as $l\left(L\right)=1$, $l\left(M\right)=2$, and $l\left(H\right)=3$, and a combined index was defined as $I=l\left(S\right)+l\left(O\right)+l\left(D\right)$, where $I\in \left\{\mathrm{3,9}\right\}$. Output risk labels were mapped as follows: $I\in \left\{\mathrm{3,4}\right\}\to VL$, $I=5\to L$, $I=6\to M$, $I=7\to H$, and $I\in \left\{\mathrm{8,9}\right\}\to VH$. This generation rule ensures interpretability and reproducibility while guaranteeing that the output risk does not decrease when any input increases [16,17].

2.4.4. Mamdani Max–Min Inference and Centroid Defuzzification

For each failure mode, membership degrees of S/O/D were calculated [13,16]. The firing strength of rule k, denoted as ${\alpha }_k$, was defined using the minimum operator as follows [14]:

$$alph{a}_k=\min \left({\mu }_{A_i}\left(S\right), {\mu }_{B_j}\left(O\right), {\mu }_{C_l}\left(D\right)\right)$$

(4)

Each rule output is represented by a clipped output membership function using ${\alpha }_k$, and overall aggregation across rules is performed by max:

$${\mu }_{Risk}^{*}\left(r\right)=\underset{k}{\mathit{max}}\left[\mathit{min}\left({\alpha }_k, {\mu }_{G_m}\left(r\right)\right)\right]$$

(5)

The final FRPN is obtained by centroid defuzzification [15,16]:

$$\mathrm{F}\mathrm{R}\mathrm{P}\mathrm{N}={\displaystyle \frac{\int r\hspace{0.17em}{\mu }_{Risk}^{\mathrm{*}}\left(r\right)\hspace{0.17em}dr}{\int {\mu }_{Risk}^{\mathrm{*}}\left(r\right)\hspace{0.17em}dr}}$$

(6)

Because numerical integration settings may differ by tool, this study ensured reproducibility by following the representative formulation above. Defuzzification was performed by discretizing r ∈ [0, 10] with Δr = 0.001 to numerically compute the centroid. Centroid defuzzification provides a stable scalar FRPN for ranking, but it also compresses the aggregated fuzzy output into a single reported value; therefore, the present results do not preserve the full dispersion of expert-related uncertainty at the reporting stage. Accordingly, the present comparison focuses on transparent rank interpretation under representative scores rather than interval-valued uncertainty propagation.

Finally, the overall Fuzzy-FMEA procedure is summarized in Figure 2.

3. Results and Discussion

3.1. Identified Hazards and Derived Failure Modes

Within the process scope, hazards were structured into three areas: the rotary kiln/activation section, where high temperature, reaction, and material transfer are coupled; the acid/base section involving corrosive chemicals; and the atmosphere/control section where off-gas leakage, overpressure, oxygen deficiency, flashback, and instrumentation/interlock failures may occur [24,25,26].

A process flow diagram is presented in Figure 3, and the derived failure modes are listed in

Table 3. Process nodes, main causes, potential effects, and current controls by failure mode.

ID	Process Node	Failure Mode	Main Causes (Examples)	Potential Effects (Examples)	Current Controls (Examples)
FM-K1	Product/dust handling and dust collection	Activated carbon dust explosion	Dust accumulation, electrostatic/friction ignition	Explosion, fire, severe injury, equipment damage	Differential pressure (DP) monitoring, cleaning, grounding
FM-K2	Rotary kiln (carbonization/activation)	Temperature control failure	Sensor drift, controller failure	Overheating, thermal runaway, refractory damage	Temperature monitoring, alarms, interlock
FM-K3	Cooling/discharge	Incomplete cooling/air ingress	Poor sealing, insufficient purge	Oxidation of hot product/localized fire	Discharge temperature monitoring, sealing inspection
FM-K4	Off-gas treatment	Reduced treatment performance	Filter damage, reduced scrubbing performance	Emission of hazardous/flammable gases	DP/flow/temperature monitoring
FM-K5	Activation gas line	Deviation in flow/composition	Valve sticking, flowmeter failure	Reaction deviation, increased CO/H2	Flow/pressure monitoring
FM-K6	Conveying/rotating parts	Blockage/residence time deviation	Conveying blockage, increased torque	Overheating/quality degradation/local ignition	Torque/current monitoring
FM-C1	Acid storage/transfer	Acid leakage	Flange/hose damage	Burns/corrosion/environmental contamination	Containment curb/tray, inspection
FM-C2	Alkali storage/transfer	Alkali leakage	Valve packing degradation	Burns/corrosion/environmental contamination	Containment/inspection
FM-C3	Dilution/mixing	Exothermic reaction/splattering	Water-acid mixing error	Splash burns/vapor generation	Work procedure, personal protective equipment (PPE)
FM-C4	Neutralization tank	Abnormal neutralization/overflow	Poor pH control	Overflow/gas generation	pH monitoring, level management
FM-C5	Piping/tank	Corrosion perforation	Inappropriate materials, accumulated corrosion	Large leakage/secondary accidents	Thickness measurement, material selection
FM-C6	Wastewater	Discharge pH limit deviation	Poor control	Regulatory violations/complaints	pH interlock
FM-A1	Duct/off-gas	Off-gas leakage (CO, etc.)	Sealing/duct damage	Poisoning/explosive mixture formation	Gas detection, ventilation
FM-A2	Purge/inerting	Oxygen deficiency	Insufficient N2 purge, insufficient ventilation	Asphyxiation risk	O2 monitoring
FM-A3	Duct/damper	Overpressure due to blockage	Damper sticking, dust accumulation	Overpressure rupture	Pressure monitoring, relief
FM-A4	Burner/ignition	Flashback	Improper air ratio	Fire/explosion	Flame monitoring, shutoff
FM-A5	Instrumentation/interlock	Instrumentation/interlock failure	Logic error/failure	Accident escalation	SIS/interlock testing
FM-A6	Ignition source control	Failure to control electrostatic ignition sources	Poor grounding	Ignition of dust/gas	Grounding, humidity, housekeeping management

. The 18 failure modes were configured to represent combined hazards rather than isolated hazards. In the rotary kiln and activation category, the interaction between high-temperature operation, dust accumulation, and air ingress is particularly critical. In contrast, within the atmosphere and control category, escalation may accelerate when off-gas containing combustible components (e.g., CO), leakage pathways, and weaknesses in detection or shutdown barriers occur simultaneously [24,25,26,27].

From a process flow perspective (Figure 3), the off-gas generated from the rotary kiln passes through a long train consisting of dust collection, afterburning/oxidation, scrubbing, and discharge, with many joints (flanges/dampers/ducts) that disperse vulnerabilities such as leakage, blockage, and differential pressure fluctuations [26]. Effective safety management therefore not only requires preventing individual equipment failures, but also maintaining overall off-gas system integrity, including sealing performance, pressure control, and ventilation, as well as ensuring the reliability of the instrumentation and alarm–interlock system throughout the process [23,24,25,26].

3.2. Classical FMEA (RPN) Results and Interpretation

Classical FMEA results identify activated carbon dust explosion (RPN = 405), rotary kiln temperature control failure (RPN = 378), and off-gas leakage (RPN = 324) as the highest-ranked risks (

Table 4. FMEA (RPN) and Fuzzy-FMEA (FRPN) results for 18 failure modes.

ID	Category	Failure Mode	S	O	D	RPN	RPN Rank	FRPN	FRPN Rank
FM-K1	Rotary kiln/activation	Activated carbon dust explosion	9	5	9	405	1	9.332	1
FM-K2	Rotary kiln/activation	Temperature control failure	9	6	7	378	2	8.000	5
FM-A1	Atmosphere/control	Off-gas leakage (CO, etc.)	9	4	9	324	3	9.221	2
FM-A5	Atmosphere/control	Instrumentation/interlock failure	8	4	8	256	4	9.221	2
FM-A6	Atmosphere/control	Failure to control electrostatic ignition sources	8	4	8	256	4	9.221	2
FM-A3	Atmosphere/control	Overpressure due to blockage	8	4	7	224	6	8.000	5
FM-A2	Atmosphere/control	Oxygen deficiency (inerting/purge)	9	3	8	216	7	7.389	7
FM-A4	Atmosphere/control	Flashback	9	3	8	216	7	7.389	7
FM-K5	Rotary kiln/activation	Deviation in activation gas flow/composition	7	5	6	210	9	6.249	13
FM-C3	Acid/base	Exothermic reaction/splattering during dilution/mixing	7	5	6	210	9	6.249	13
FM-K4	Rotary kiln/activation	Reduced off-gas treatment performance	8	5	5	200	11	7.388	12
FM-C1	Acid/base	Acid leakage (storage/transfer)	8	4	6	192	12	7.389	7
FM-C2	Acid/base	Alkali leakage (storage/transfer)	8	4	6	192	12	7.389	7
FM-C5	Acid/base	Corrosion perforation	8	3	8	192	12	7.389	7
FM-K6	Rotary kiln/activation	Blockage/residence time deviation	7	4	6	168	15	6.249	13
FM-C4	Acid/base	Abnormal neutralization/overflow	7	4	6	168	15	6.249	13
FM-K3	Rotary kiln/activation	Incomplete cooling/air ingress	6	5	4	120	17	4.839	17
FM-C6	Acid/base	Discharge pH limit deviation	6	4	5	120	17	4.839	17

). These findings indicate that high-severity events, combined with weak detectability (high D), dominate the priority structure [10]. Even when occurrence (O) is assessed as moderate, dust explosion risk rises sharply when detectability is poor, reflecting process characteristics in which dust accumulation, electrostatic hazards, and ignition sources are difficult to detect perfectly in advance and escalation can be rapid [27]. In addition, many tie scores occurred in RPN calculations, making it difficult to subdivide mid-risk groups. For example, leakage and corrosion perforation in the acid/base domain may be tied by RPN, but the former requires short-term event-centered management, whereas the latter is driven by long-term degradation management; thus, improvement strategies differ. Therefore, classical RPN is useful for initial screening of top risks, but is limited in supporting decision-making within tied ranks [10,11,12].

3.3. Fuzzy-FMEA (FRPN) Results and Interpretation

Fuzzy-FMEA results identified dust explosion (FRPN = 9.332), off-gas leakage (FRPN = 9.221), instrumentation or interlock failure (FRPN = 9.221), and failure to control electrostatic ignition sources (FRPN = 9.221) as the highest-ranked risks (Table 4). The comparison between RPN_norm and FRPN in Figure 4 indicates that the top risk group is consistently high in both methods, but FRPN more clearly elevates barrier-failure-type risks (instrumentation/interlock and ignition source control) into the top tier. This can be understood as a result of fuzzy rule-based integration that emphasizes the managerial meaning of vulnerability of D and barrier failures, thereby revealing the risk characteristics that may not be fully captured by a simple multiplicative index. This behavior is structurally embedded in the rule base: because the output label is determined by the ordinal sum $I=l\left(S\right)+l\left(O\right)+l\left(D\right)$, the VH region expands rapidly as $D$ increases. In Figure 5, the number of $\left(S,O\right)$ combinations mapped to VH increases from 0/9 at $D=L$ to 1/9 at $D=M$ and 3/9 at $D=H$ (while the number mapped to $\ge H$ increases from 1/9 to 3/9 to 6/9), explaining why barrier-failure-type items with high $D$ are promoted to the top tier.

Figure 5 summarizes the output structure of the 27-rule fuzzy rule base as a heatmap by D level, visualizing how combinations of input linguistic terms (S/O/D) map to output labels (VL-VH) and how monotonicity and vulnerability of D are implemented at the rule level. The rule base summary shows a clear tendency for the output risk label to increase as D increases. In systems such as activated carbon processes, where detection and shutdown barriers are critical for accident prevention, this structure is practically meaningful. Meanwhile, because input linguistic terms were simplified into three levels, some regions may yield identical FRPN values, suggesting a potential limitation and a possible extension (e.g., increasing input linguistic granularity and expanding the rule base) to further improve mid-risk resolution. Accordingly, the Fuzzy-FMEA proposed here should be interpreted not as inventing a completely new ranking, but as a decision-support tool that more clearly highlights the risks in which detection vulnerability and barrier failures are coupled, strengthening the rationale for selecting engineering alternatives. In this sense, Figure 4 and Figure 5 provide graphical views of how the fuzzy system redistributes risk scores relative to classical RPN and how high-D vulnerability expands the upper-risk region at the rule level.

3.4. Meaning of the Differences Between RPN and FRPN Rankings

Overall, RPN and FRPN rankings showed high consistency (Spearman ρ = 0.871; Kendall τ = 0.752), and the Top 5 set was identical in both methods (

Table 5. Rank comparison and rank-correlation analysis (RPN vs. FRPN).

Metric	Value
Spearman rank correlation (ρ)	0.871
Kendall rank correlation (τ)	0.752
Top 5 overlap	5/5
Top 10 overlap	8/10
Mean |Δrank|	2.06
Max |Δrank|	5

). Nevertheless, mid-risk items exhibited non-negligible rank rearrangements (mean |Δrank| = 2.06; max |Δrank| = 5), reflecting analytical reordering across clusters that are tied or compressed under the multiplicative RPN. Notably, the acid/base leakage and corrosion perforation cases (FM-C1, FM-C2, and FM-C5) increased from RPN rank 12 to FRPN rank 7, whereas the activation-gas deviation and dilution/mixing exothermic reaction cases (FM-K5 and FM-C3) decreased from RPN rank 9 to FRPN rank 13. These shifts support linking improvement strategies not only to event–response actions, but also to barrier reliability and detectability enhancement, contributing to clearer investment prioritization from a process safety perspective. Furthermore, the core of this study is not a superiority comparison between RPN and FRPN, but rather the explicit interpretation of transfer pathways for top-risk scenarios and the presentation of actionable engineering-alternative packages for high-priority items. These analytical differences become practically meaningful only when translated into scenario-based engineering decisions; accordingly, Section 3.5 moves from comparative ranking behavior to the construction of prioritized improvement packages.

3.5. Category-Wise Risk Characteristics and Priority Improvement Measures

To discuss engineering risk-reduction directions for a rotary kiln-based activated carbon manufacturing process, this section interprets the process characteristics and escalation mechanisms, focusing on the top five failure modes identified by Fuzzy-FMEA, and presents actionable engineering alternatives. The top risks in this process are characterized not by single equipment failures, but by rapid escalation when high-temperature operation, dust accumulation, flammable/toxic off-gas, potential air (oxygen) ingress, sealing/duct integrity, and instrumentation/interlock reliability are coupled. Accordingly, risk-reduction measures should be structured as a package from the perspectives of (i) prevention to reduce the likelihood of causes (occurrence suppression), (ii) detection and shutdown to identify abnormal states early and cut escalation before transfer (barrier reliability), and (iii) mitigation/isolation to limit consequences and escalation scale if an incident occurs.

Table 6. Escalation characteristics and priority improvement packages for top five risks (based on FRPN; P = prevention/occurrence reduction, D = detection/shutdown enhancement to reduce detectability vulnerability, and M = mitigation or escalation-consequence reduction).

ID	Category	Failure Mode	Escalation Mechanism	Key Vulnerable Barrier	Priority Improvement Package
FM-K1	Rotary kiln/activation	Activated carbon dust explosion	Dust accumulation (collection/transfer/storage) + ignition sources (electrostatic discharge, friction, overheating) + oxygen ingress → rapid transition to explosion/fire	Dust is difficult to fully detect in advance, and escalation is rapid (high D vulnerability). Housekeeping/DP monitoring may be indirect indicators.	P (↓O): Minimize dead zones/pockets that promote dust holdup; improve cleaning access; design to reduce dust leakage; standardize vulnerable sections. D (↓D): Refine DP trend alarms; use motor current/torque and bearing temperature for early diagnostics and maintenance triggers. M (↓S/escalation cut): Consider explosion isolation and mitigation (venting/compartmentation) where feasible; strengthen ignition control (systematic grounding/bonding).
FM-A1	Atmosphere/control	Off-gas leakage (CO, etc.)	Multiple leakage points along long ducts/seals/flanges/dampers → leakage → worker poisoning + (conditional) explosive mixture formation → ignition → accident escalation	Leakage points are dispersed, increasing management difficulty. If detect-alarm-shutdown remains only as alarms, escalation is hard to cut.	P (↓O): Standardize specs and inspection criteria for ducts/seals/joints; preventive replacement of degraded sections; focused inspection/leak management of vulnerable parts (dampers/expansion joints). D (↓D): Place CO (or key component) detectors at vulnerable points with calibration; monitor negative pressure/ventilation status. M (escalation cut): Link detection signals to ventilation reinforcement or safe shutdown (fuel cut-off/shutoff valves, etc.) to sever leak-accumulation-ignition.
FM-A5	Atmosphere/control	Instrumentation/interlock failure (barrier failure)	Failure of detection-alarm-shutdown under abnormal states → accident escalation (amplifier of other risks)	Barrier failure is a common amplification factor for the entire process; bypass, insufficient testing, and poor MOC worsen D.	P (↓O): Redundancy and cross-check diagnostics for critical instrumentation; strengthen logic change MOC; improve alarm quality via alarm rationalization. D (↓D): Set and comply with proof test intervals; establish bypass control including approval/duration/return confirmation. M (escalation cut): Redefine functions from an IPL perspective as shutdown, not merely alarm, where necessary.
FM-A6	Atmosphere/control	Failure to control electrostatic ignition sources	Poor grounding/bonding/non-conductive parts + dust/gas presence → electrostatic discharge ignition → rapid transition to explosion/fire	Ignition control becomes fatal when coupled with dust/gas hazards; field management may fragment into partial grounding.	P (↓O): Standardize grounding/bonding design (including mobile equipment); clarify criteria for electrostatic-vulnerable materials (hoses/gaskets); reduce friction sources (alignment/bearing condition). D (↓D): Periodic continuity/resistance checks; checklist-based controls during work/maintenance. M (escalation cut): Integrate with (K1) explosion isolation/mitigation measures as a package.
FM-K2	Rotary kiln/activation	Temperature control failure	Sensor drift/controller or fuel-system anomalies/operating deviations → overheating, thermal runaway, and refractory damage → fire/process instability	Monitoring and alarms alone are insufficient for pre-escalation shutdown; without Hi-Hi shutdown/IPL, D becomes vulnerable.	P (↓O): Strengthen strategies for operating condition control (fuel/air ratio, etc.) and control system health. D (↓D): Temperature measurement redundancy + cross-check diagnostics; automatic fuel cut-off via Hi-Hi interlocks; interlock proof test and bypass control. M (escalation cut): Secure intervention time via ROC-based early warning; refine safe shutdown sequences from an IPL perspective where necessary.

summarizes the escalation mechanisms, vulnerable barriers, and prioritized engineering alternatives for the top five FRPN scenarios. At the top five cutoff, FM-K2 and FM-A3 had identical FRPN values (8.000); to maintain a fixed set of five representative scenarios, FM-K2 was selected as the tie-breaker item based on its higher RPN.

First, activated carbon dust explosion (FM-K1) can act as a representative trigger of catastrophic escalation in activated carbon manufacturing, and thus requires the highest-priority engineering response among the top risks [27,28]. Dust explosion risk increases sharply when dust accumulates in dust collectors, conveyors, and storage/handling areas, ignition sources such as electrostatic discharge, friction, and overheating exist, and oxygen is supplied through air ingress [27]. Engineering alternatives should prioritize design improvements that structurally reduce dust accumulation (minimizing dead zones/pockets that promote dust holdup, improving reservoir/pocket structures, securing cleaning accessibility, and standardizing sections prone to dust buildup) [29]. In addition, operating data, such as dust collector DP trends and conveying motor torque/current and bearing temperature, can provide precursor indicators of accumulation/friction, enabling early-warning alarms and maintenance triggers that increase opportunities for intervention before escalation. Finally, ignition-source control should be reinforced through systematic grounding/bonding and management of non-conductive parts/hoses; where feasible, explosion isolation and mitigation designs (e.g., compartmentation) should also be considered [30,31,32,33].

Second, off-gas leakage (FM-A1) is related to structural vulnerabilities in which off-gas generated from the rotary kiln passes through long systems connected by dust collection–afterburning/oxidation–scrubbing, and leakage may occur at many joints such as ducts, seals, flanges, and dampers. If the off-gas contains CO or other hazardous/flammable components, leakage can not only lead to worker poisoning, but also to the formation of explosive mixtures, necessitating a packaged engineering response [26]. Preventive measures include standardizing specifications and inspection criteria for ducts/seals/joints, preventive replacement of degraded sections, and focused inspection of leakage-prone parts (dampers, expansion joints, etc.). Next, to prevent leakage from escalating into hazards, negative-pressure maintenance of the system, strengthened local exhaust/ventilation, and gas detector placement (e.g., CO) with calibration programs should be established to achieve early detection and exposure reduction simultaneously [26]. In addition, configuring logic such that detection signals do not only remain as alarms, but are linked to ventilation enhancement or safe shutdown (e.g., shutoff valves and fuel cut-off) can sever the leak–accumulation–ignition escalation pathway [23,24,25].

Third, instrumentation/interlock failure (FM-A5) represents barrier reliability issues that commonly appear in the top risk group [24,25]. In coupled high-temperature-gas-dust systems such as activated carbon processes, accident escalation is determined less by stability during normal operation than by whether detection–alarm–shutdown functions work properly during abnormal state transitions [9,24]. Therefore, engineering alternatives should include: appropriate redundancy and cross-check diagnostics for critical measurements; improved alarm quality (false/missed alarm management); strengthened management of change (MOC) for logic modifications; the establishment and compliance of proof–test intervals; and strict bypass control systems, including approval, duration control, and confirmation of restoration [22,23,24,25]. In particular, recommended items in the manuscript (e.g., SIS/interlock testing) only improve detection-and-shutdown performance when implemented in actual operation and maintenance systems and constitute key levers for meaningfully lowering the FRPN of top risks [24,25].

Fourth, failure to control electrostatic ignition sources (FM-A6) is a critical engineering risk directly linked to the ignition of dust explosions and flammable gases [34]. Electrostatic hazards can increase due to poor grounding/bonding, use of non-conductive conveying parts/hoses, and increased friction sources from rotating components; escalation is promoted when combined with dust accumulation or leaked gases [35,36,37,38]. Engineering alternatives should prioritize standardizing grounding/bonding designs (including mobile equipment) and establishing periodic verification programs (continuity/resistance measurements) [34,35]. In addition, clear management criteria should be set for electrostatically vulnerable materials and parts (hoses, gaskets, etc.), and friction sources in rotating components (alignment, bearing condition, overheating) should be managed as operating indicators to reduce ignition likelihood. Ultimately, electrostatic control should be emphasized not merely as a sub-item of dust explosion countermeasures, but as an independent engineering alternative that severs escalation mechanisms of top risks [34,37].

Fifth, temperature control failure (FM-K2) is a key risk in rotary kiln processes because it can escalate into overheating, thermal runaway, and refractory damage under high-temperature operation, and it is also ranked high under classical RPN [7,20]. Since temperature control failure can arise from sensor drift, controller errors, fuel system anomalies, and deviations in operating conditions, engineering design from an Independent Protection Layer (IPL) perspective is required beyond monitoring and alarms [9,24,25]. Representative measures include configuring interlocks so that automatic fuel cut-off occurs under high–high-temperature (Hi-Hi) conditions, implementing redundancy and cross-check diagnostics for temperature measurements to detect sensor abnormalities early, and systematizing proof tests and bypass controls for interlock logic [22,23,24,25]. Introducing rate-of-change (ROC)-based early warnings can also extend the time available for intervention before escalation, contributing to both process stability and safety.

In summary, the top five risks can be interpreted as results of coupled structural vulnerabilities across the process (dust handling, long off-gas systems, high-temperature operation) and barrier reliability (instrumentation/interlocks and ignition source control). Therefore, it is reasonable to prioritize risk-reduction strategies that: (i) manage dust inventory and ignition sources simultaneously, (ii) establish packaged integrity/ventilation/detection measures for off-gas systems, (iii) ensure barrier reliability through interlocks/proof testing/bypass control, and (iv) block escalation of temperature control failures through independent protection layers centered on fuel cut-off. Although acid/base items did not enter the top five FRPN set, the upward shift in FM-C1, FM-C2, and FM-C5 to FRPN rank 7 indicates that leakage and corrosion management remain the most important secondary improvement cluster outside the top tier.

4. Conclusions

This study examines a rotary kiln-based activated carbon manufacturing process and structured its hazards across three domains: rotary kiln and activation operation, acid/base handling, and process atmosphere and control. Classical FMEA and Fuzzy-FMEA were applied in parallel to compare and interpret risk prioritization results. Classical FMEA identified dust explosion, temperature control failure, and off-gas leakage as the dominant risks, confirming that activated carbon manufacturing represents a complex, high-hazard system in which high temperature, gas, and dust interactions are closely coupled. Fuzzy-FMEA preserved this core high-risk group while more clearly elevating barrier-related factors, particularly instrumentation or interlock failure and inadequate electrostatic ignition control, into the top tier. This provides additional interpretive insight by emphasizing barrier reliability in risk prioritization. Because the same expert panel and final S/O/D set were used for both methods, the comparison should be interpreted as a within-panel methodological comparison under a common evidence base.

The RPN and FRPN rankings showed strong consistency (Spearman’s ρ = 0.871; Kendall’s τ = 0.752), and the top five failure modes were identical under both approaches. However, rank rearrangements within the mid-risk group improved differentiation among previously tied RPN values, thereby enhancing resolution for resource allocation decisions. Based on these findings, risk reduction in rotary kiln-based activated carbon manufacturing should prioritize the following:

(i)

Dust explosion prevention, including dust collection, electrostatic control, and isolation measures;

(ii)

Off-gas leakage monitoring and system integrity management;

(iii)

Functional reliability of instrumentation and interlock systems, including verification, proof testing, and bypass control;

(iv)

Temperature control protection through independent protection layers, high–high shutdown logic, and early-warning diagnostics;

(v)

Acid/base leakage and corrosion management as the leading secondary risk cluster outside the top-tier FRPN group, supported by material selection, thickness monitoring, and inspection programs.

Future research should further improve precision in process safety decision-making by (i) retaining expert-level dispersion or interval-valued uncertainty rather than reporting only scalar defuzzified FRPN values, (ii) expanding rule bases to increase mid-risk resolution, (iii) separating rules by accident scenario, and (iv) integrating the framework with protection-layer performance.