The Ishikawa diagram, also known as the cause-and-effect diagram, is a quality management tool developed by Kaoru Ishikawa in the 1950s [19][20]. We will use this tool upstream to comprehensively identify the causes of a critical failure of an HFO separator, and then perform the FMECA analysis in accordance with IEC 60812 [21]. For each failure mode, we will assign:

severity (S): from 1 (minor: production stoppage of less than 15 minutes) to 10 (critical: power plant shutdown of 2 hours or more);occurrence (O): from 1 (rare failure) to 10 (very frequent failure: at least once per week);detection (D): from 1 (certain detection, obvious signs) to 10 (impossible to detect, no signs);criticality (C): C = S × O × D .

MTBF (Mean Time Between Failures): Mean time between two successive failures.For each separator MTBF = Operating time − ∑ Failure hours Total failure count (1)MTTR (Mean Time To Repair): mean repair time (hours).Operational availability: D = MTBF MTBF + MTTR (2)Failure rate: λ = 1 MTBF (3)

The reliability function R ( t ) was estimated using the Kaplan-Meier method [22], with right-censoring applied, since, as stated in the introduction, one of the three separators was undergoing corrective maintenance while the other two were still in perfect working order; their future failure times are therefore unknown. This corresponds to a multiple-failure scenario with renewal, as the separators fail and are subsequently repaired. Accordingly, we consider inter-failure times rather than absolute age. In this context, right-censoring applies to the last interval of each separator:

S1 has its last interval censored (as it was under maintenance);S2 and S3 have their last intervals complete (since they failed before the end of the study).

The Kaplan-Meier estimator is the standard reference tool in reliability analysis and provides a more accurate estimate of the survival probability of the equipment or the system.

where t i are the observed failure times, d i is the number of failures at time t i , n i is the number of subjects “at risk” just before t i .

We summarize the process for calculating reliability indicators and analyzing the results in the algorithm shown in Figure 1 below. A computation code was developed using Matlab version R2020a for data input, execution of algorithm commands, and curve plotting.

Figure 1.Calculation algorithm.

Power plant characteristics

The thermal power plant is located in the industrial zone of the Kossodo district, north of the city of Ouagadougou. It consists of eight (08) generator sets, one (01) of which has been out of service since the beginning of 2019. These generator sets comprise diesel engines and alternators mechanically coupled for the conversion of mechanical energy into electrical energy. Since 2021, the installed capacity has been approximately 112.7 MW, and its production covers 73.86% of the Ouagadougou Regional Consumption Center (CRCO). Table 1 summarizes the characteristics of the power plant’s machinery.

Table 1.Characteristics of the generator sets and construction stages of the Kossodo power plant.

Ouaga East comprises the last three generator sets (G9, G10, and G11), whose separators are the subject of this study.

Centrifugal separator characteristics

Three GEA centrifugal separators are studied in this work. They are arranged in parallel and are not operated simultaneously except during periods of high electricity production during hot weather (March to June). Figure 2 shows a view of the HFO unloading area (Figure 2(a)) and an image of one of the installation’s separators (Figure 2(b)).

The separation process in a centrifugal separator relies on the density difference between heavy fuel oil and impurities (water, solid particles, etc.). The heavy fuel oil is introduced into the separator through the inlet piping. The mixture of fuel and impurities is directed toward the rotating bowl. This rotation generates a centrifugal force that projects the impurities toward the bowl walls. The fuel, being less dense, remains closer to the center of the bowl. The separated impurities are discharged via the sludge discharge system. This process ensures that only clean, impurity-free fuel is used in the engines or boilers.

Figure 2.(a) External view of the unloading area at the Ouaga East thermal power plant; (b) HFO separator of the Ouaga East thermal power plant.

HFO: the properties are measured using an Anton Paar DMA 4200 M apparatus, which offers a precision of 0.0001 g/cm³ and an operating temperature of up to 150˚C. The values of the constants measured by the operators are presented in Table 2.

Table 2.Physical Properties of the HFO (Sample 178_SPTN-0169F703/6294F703(1)_178).

Commissioning date of the Ouaga East power plant: September 1, 2021.Observation period: September 8, 2021 to August 19, 2024.Sources:maintenance logs, logbooks, intervention reports, fault sheets.Data collection: The data extracted from the aforementioned sources form the monitoring dashboard used to track the operation of the separators. For each separator, an Excel spreadsheet logs the date and duration of each interruption, service hours, operating time, failed components (including both individual and simultaneous failures), failure type, corrective actions, as well as parts and labor costs. We subsequently propose a stepwise algorithm to process these data.

We applied the Ishikawa diagram to the data collected on the HFO separators during the observation period. This allowed us to determine the origin of the failures. Figure 3 summarizes the results obtained.

From the analysis of this diagram, it appears that equipment is the main source of failures, accounting for 33% of the causes. Method and labor each represent 20% of the causes, material 14%, and environment 13%. We will therefore primarily focus our failure mode, effects, and criticality analysis on the centrifugal separators themselves.

Figure 3.Ishikawa diagram of the HFO separators.

We now apply the FMECA to all three HFO centrifugal separators. The results are presented in Table 3 below.

Table 3.FMECA results for all three separators.

Based on the elementary individual and simultaneous data gathered on the operating behavior of the three separators (S1, S2, and S3) over the observation period, and applying formulas (1) through (3), the resulting reliability indicator values are compiled in Table 4 below.

Availability values are identical, since they depend on the overall cumulative figures (i.e., the same proportion of downtime).

Table 4.Reliability indicator values by separator.

Based on the recorded data pertaining to simultaneous failures across the separators, the histogram of combinations illustrated in Figure 4 is established.

Figure 4.Combination histogram of the three HFO separators.

Concurrent failures of two separators (S1 + S2) or three separators (S1 + S2 + S3) are observed to be frequent (approximately 11 occurrences over a 4-year period). This points to underlying external factors:

Ambient temperatures are extremely high from March to June. The collected data indicate a higher number of interventions during this period.HFO quality (viscosity, water content, impurities) can affect all separators operating in parallel.

The influence of these factors is indeed demonstrated in [14] and [15].

Table 5 shows the aggregated data derived from the collection over the observation period for the three HFO centrifugal separators at the Ouaga Est (Kossodo) power plant.

Table 5.Aggregated data across the three centrifugal separators (pooled).

The pooled global indicators for the three separators are thus:

Overall failure rate: λ = 0.01159 failures/hours. This corresponds to one failure every 86 hours.Overall MTBF: 1 / λ = 86.28 hours. i.e., 3.6 days.Overall availability (maintenance excluded): D = 69.72 % .Unplanned stoppage rate: 1 − D = 30.28 % . Corresponding to 30.3% of the operating time spent in failure.Global median time to failure: 548.3 hours.

From these results, it emerges that the Ouaga Est (Kossodo) thermal power plant has low availability ( D ≺ 70 % ), with unplanned downtime reaching a very high level (2060 hours out of 4745 operating hours). Corrective maintenance is the predominant maintenance strategy.

Based on the actual individual and combined data from the separators, and by applying Equation (4) above, the Matlab computation yields the curves shown in Figure 5 and Figure 6. Figure 5 presents the estimated survival function for each of the three separators. The median survival time for separator S1 is 281 h; for S2 it is 1362 h; and for S3 it is 2400 h. The differences among S1, S2, and S3 can be explained by uneven usage during the observation period. S1 was used much more frequently, hence experienced more failures and a lower median. S3 was subjected to little demand, resulting in fewer failures. Figure 6 shows the estimated survival function for all three separators combined, with a 95% confidence interval. A rapid decrease in the reliability function is observed after 209.5 hours of continuous operation, with a probability of proper functioning of 0.6. This probability drops to 0.25 after 1460 hours, i.e., 60 days of operation. At this point, a slope break appears, suggesting premature wear of bearings and seals. This result was predicted by the FMECA analysis performed earlier and is confirmed by the findings reported in [14][23]-[25]. At 6000 hours, a decrease in the failure frequency is observed, marked by a horizontal plateau. The system enters a wear‑out phase before 15,000 hours of operation. The overall median lifetime is 548.3 hours.

Figure 5.Survival curves by individual separator.

Figure 6.Global reliability curve with 95% confidence bounds.

The mean cost of a single corrective intervention is 700,000 CFA francs (approximately 1200 USD), including spare parts and labor. With a total of 55 failures recorded, the average cumulative repair cost over the observation period reaches 38,500,000 CFA francs (i.e., 64,200 USD). Considering the cost per ungenerated kWh, the production loss attributable to unscheduled downtime is estimated at 87,000 USD.

The low MTBF (86.28 h) compared to standard values can be attributed to several external and context‑specific factors related to the city of Ouagadougou:

High ambient temperature: during hot periods (March to June) in Ouagadougou, peak temperatures sometimes reach 43˚C. Above 40˚C, the viscosity of HFO decreases, which alters the hydraulic load on the discs and accelerates bearing wear.Variable HFO quality: water content occasionally exceeding 1.2%, together with the presence of fine sand (not captured by coarse filters), accelerates clogging.Predominantly corrective maintenance: lack of continuous vibration and thermal monitoring. Indeed, the Mean Time To Repair (MTTR) is 37.47 h (1.56 days), with an unplanned downtime‑to‑maintenance ratio of 12.90.Non‑certified spare parts: use of bearings of questionable origin and lower quality.

This study has several limitations, which relate to:

Manual data recording: certain maintenance interventions were incompletely documented (under-reporting).Undifferentiated failure modes: no distinction was made between independent failures and induced failures (e.g., clogging can cause bearing overheating).Absence of on-site vibration data: no measurements of in-situ vibratory stresses were available.

In light of the results presented above, we propose the following recommendations for industrial operators:

Enhance preventive maintenance through regular disc cleaning, HFO quality control, and vibration monitoring.Mitigate concurrent failures by implementing preheating or HFO conditioning (upstream separator, clarification centrifuge).Pay particular attention to high-temperature periods (March to June).Provide operator training for early fault detection (e.g., abnormal noise or temperature) and for minimizing setting errors.Employ only certified replacement parts (SKF, FAG).Monitor repair durations to identify process bottlenecks.