Heavy Fuel Oil Quality Dependence on Blend Composition, Hydrocracker Conversion, and Petroleum Basket

Abstract

The production of very-low-sulfur residual fuel oil is a great challenge for modern petroleum refining because of the instability issues caused by blending incompatible relatively high-sulfur residual oils and ultra-low-sulfur light distillates. Another obstacle in the production of very-low-sulfur residual fuel oil using hydroprocessing technology is the contradiction of hydrodesulfurization with hydrodemetallization, as well as the hydrodeasphaltization functions of the catalytic system used. Therefore, the production of very-low-sulfur residual fuel oil by employing hydroprocessing could be achieved by finding an appropriate residual oil to be hydroprocessed and optimal operating conditions and by controlling catalyst system condition management. In the current study, data on the characteristics of 120 samples of heavy fuel oils produced regularly over a period of 10 years from a high-complexity refinery utilizing H–oil vacuum residue hydrocrackers in its processing scheme, the crude oils refined during their production, the recipes of the heavy fuel oils, and the level of H–oil vacuum residue conversion have been analyzed by using intercriteria and regression analyses. Artificial neural network models were developed to predict the characteristics of hydrocracked vacuum residues, the main component for the production of heavy fuel oil. It was found that stable very-low-sulfur residual fuel oil can be manufactured from crude oils whose sulfur content is no higher than 0.9 wt.% by using ebullated bed hydrocracking technology. The diluents used to reduce residue viscosity were highly aromatic FCC gas oils, and the hydrodemetallization rate was higher than 93%.

1. Introduction

Heavy oil, shale oil and gas, hydrates, and other resources will inevitably be important replacements for fossil fuels in the future [1]. Heavy fuel oil (HFO) is a low-cost energy resource used in marine transport [2,3,4,5] and in diverse industrial applications [5,6,7]. It is a complex mixture typically produced in petroleum refineries by blending residual oils obtained from vacuum residue visbreaking [8,9,10], vacuum residue desulfurization [11,12,13], vacuum residue hydrocracking [14,15], asphalt from deasphaltization, and pyrolysis resin [16,17] with cutter stocks such as fluid catalytic cracking gas oils (light cycle oil, heavy cycle oil, and slurry oil), kerosene, primary and secondary diesel oils, and vacuum gas oils [18,19,20]. Heavy fuel oil may also contain various bio-additives [21,22]. Its current demand worldwide amounts to two million barrels per day [23]. The stringent environmental regulations targeting a reduction in sulfur oxides (SOxs), nitrogen oxides (NOxs), and particulate matter (PM) [24] and the increased global production of heavy sour crude oils have increased the importance of residue hydrocracking [25,26,27]. It not only reduces the level of contaminants in petroleum residue but also enhances its conversion to high-value automotive fuels and feeds for petrochemistry [28,29,30]. Despite the reduction in impurities in hydrocracked residual oils compared to visbreaked (thermally converted) heavy oils, the colloidal instability of the hydrocracked residues obtained from certain crude oils and its dependence on reaction severity are still pending issues [31,32,33]. The colloidal stability and the related sedimentation of residual marine fuels have been subject of several investigations [33,34,35,36,37,38,39,40,41]. Kuzmin et al. [34] investigated eight laboratory-prepared residue marine fuels which contained between 0 and 30% waste cooking oil (WCO) and established that the addition of WCO to residue marine fuels (RMFs) decreased the total sediment accelerated (TSA) rate from 0.044 to 0.027 wt.%. Sultanbekov et al. [35] studied three residue marine fuels (RMD–80 (density at 15 °C (D₁₅) of 901 kg/m³); RMG–380 (D₁₅ of 956 kg/m³); and RMK–700 (D₁₅ of 976 kg/m³)) which were mixed with a secondary oil (D₁₅ of 833.5 kg/m³) that did not contain asphaltenes. They found that when the n-paraffin content in the studied RMF samples was lower than 57 wt.%, then all 112 studied RMF samples were stable according to their total sediment potential (TSP). They also determined that the stability of RMFs decreased (TSP increased) with the enhancement in the asphaltene content in the range 0.5–3.5 wt.%. Sultanbekov et al. [36] developed a method of determining the compatibility and stability of fuel mixture components based on laboratory tests and using machine learning methods. According to the authors, the model can be used to predict the sediment content of mixed marine residual fuels that have a desired level of sulfur. Vráblík et al. [37] prepared seven model samples of RMFs made of visbreaker residue (D₁₅ of 961.2 kg/m³; asphaltenes of 25.34 wt.%), ultra-low-sulfur diesel (D₁₅ of 835.8 kg/m³), and fluid catalytic cracking light cycle oil (FCC LCO) (D₁₅ of 956.0 kg/m³). The stability of the samples was evaluated according to the TSA rate, which varied between 0.031 and 0.098 wt.% [34]. The authors concluded that a stable RMF can be obtained when the content of the aromatic FCC SLO is maximized while the amount of paraffin–naphthene ULSD is minimized in the blended fuel. Efimov et al. [38], in their research, employed five oils (ULSD (D₁₅ of 835.8 kg/m³); FCC LCO (D₁₅ of 956.0 kg/m³); straight-run vacuum residue (D₁₅ of 990.2 kg/m³); visbreaking residue (D₁₅ of 961.2 kg/m³); and heavy pyrolysis resin (D₁₅ of 1073.3 kg/m³)) to develop a UNIFAC model to predict the stability of very-low-sulfur residual marine fuel (VLSFO). They suggested that the proposed model can be used for marine fuel blending simulations. The main parameters for the database are UNIFAC compositions of petroleum products along with the asphaltene heats of fusion [35]. Smyshlyaeva et al. [39] investigated in their study seven residual oils (straight-run vacuum residues (SRVRs), visbreaker residues, asphalt from deasphaltization, and pyrolysis resins) and two crude oils. They extracted asphaltenes from these nine oils and added them to RMG 380 marine fuel in amounts between 1 and 5 wt.% and measured TSA rate of the blends. They found that the TSA rate of RMG 380 from 0.03 wt% at zero asphaltene content exponentially increased to 3.07 wt% at 5 wt% asphaltenes derived from pyrolysis resin in the marine fuel. Even at 1 wt.% asphaltenes derived from pyrolysis resin in the blend with RMG 380, the marine fuel becomes unstable (TSA = 0.27 wt.%). The researchers found out that the stability of the fuel compositions decreased in the order asphalt > vacuum residue > crude oil > visbreaker residue > heavy pyrolysis resin. They found that the higher the atomic H/C ratio of the asphaltenes (the lower the aromaticity), the higher the sedimentation stability (the lower the TSA values) of the residual marine fuel with added asphaltenes [39]. Mitusova et al. [40] used straight-run residual fuel oil, visbreaking residue, and FCC LCO to prepare four high-viscosity samples of marine fuels and measured their xylene equivalent as an indicator of the resistance of the fuel samples to form sediments. They reported that a xylene equivalent (XE) not exceeding 25/30 vol.% is one of the criteria to prove that the residual fuel is of primary origin. While blends of straight-run fuel oil with FCC LCO are very stable (XE 20/25 vol.%), those of visbreaking residue with LCO are unstable (XE 100), with the sediment content increasing from 0.01 to 0.6 wt.%. They showed that the addition of dispersant chemicals in an amount of 0.05 vol.% can improve the xylene equivalent and reach the XE standard of 25/30 if the base fuel has an XE not higher than 30/35 vol.%. Vermeire and Heyberger [41], studying 35 fuel oil blends, found that predicting the compatibility of fuel oil blends using the three methods ASTM D 7157 [42], ASTM D 7112 [43], and ASTM D 7060 [44] showed a percentage of good predictions of measured stability using TSP and TSA of 74, 77, and 73%, respectively, when the S-value, P-value, and P-ratio were set to 1.1. These data indicate that the stability of one-quarter of the fuel blends studied, as expressed by TSP and TSA, cannot be well-predicted by two-solvent titration methods. Zhou et al. [45] using molecular dynamics (MDs) simulations investigated the effect of adding a light distillate component (hexadecane) to heavy fuel oil in order to produce low-sulfur fuel oil. MD calculations exhibited that the addition of light distillate leads to a tremendous increase in the peak height in the radial distribution function of the asphaltene–asphaltene pair, thereby facilitating aggregation and precipitate formation in the fuel oil. All of these investigations indicate that the production of stable fuel oil with a lower sulfur content is a great challenge. Different asphaltenes have been shown to contribute to sediment formation differently [39], and the asphaltene content and that of light distillates are also important [35]. This makes the prediction of heavy fuel very difficult.

Another option to produce lower sulfur residual fuel oil is to use residue hydrocracking technology with an appropriate catalytic system and suitable residual oils. As in the production of very-low-sulfur residual marine fuels by blending residues with ULSD, where incompatibility is the main issue, high-activity hydrodesulfurization (HDS) catalysts exhibit poor hydrodemetallization (HDM), hydrodeasphaltization (HDA), and sedimentation control [30]. Therefore, in the case of residue hydrocracking, the search for a way to produce residual fuel oil with lower sulfur content and high stability should be directed towards using a suitable residual oil.

Metaheuristic methods (artificial intelligence: AI) can discover the hidden principles in the data and assist in the development of models used in heavy oil conversion processes, such as fluid catalytic cracking [46] and hydrocracking [47,48,49], when the relations between input and output are hard to unravel directly [50,51]. These approaches enable us to reduce complexity and provide fast results. AI methods can be classified as artificial neural networks [47,48,49], convolutional neural networks [52], deep neural networks [53], least squares support vector regressions [54,55], dynamic partial least squares (DPLSs) and partial least squares (PLSs) [47], neural decision trees (NDTs) [56], principal component analysis [57], and ensemble learning [58]. Therefore, the metaheuristic models are appropriate for industrial applications with high complexity, and for this reason one of these methods, ANN, was employed in this research.

In the current study, the performance of a commercial ebullated bed vacuum residue H–oil hydrocracking during the processing of vacuum residual oils originating from 41 crude oils was investigated with the aim of determining the relationships between the operating conditions, the condition of the catalyst system, and the properties of the vacuum residue feed blend to the resulting hydrocracked vacuum residue and the finished fuel oil after blending with cutter stocks.

The H–oil vacuum residue conversion during the study varied between 46.5 and 92.6 wt.%. The density of the hydrocracked vacuum residue (VTB = vacuum tower bottom) fluctuated between 977.9 and 1125.9 kg/m³, and the VTB sulfur content oscillated between 0.5 and 2.2 wt.%. The density of the finished heavy fuel oil product varied between 957.4 and 1048.9 kg/m³, while its sulfur content changed between 0.7 and 2.2 wt.%. Characteristics data for 120 finished fuel oil cargoes produced over 10 years (between 2015 and 2025) were evaluated by intercriteria analysis (ICrA). Characteristics data for 140 H–oil vacuum residue blend feeds and VTBs were evaluated by ICrA, and linear and nonlinear regression techniques. Artificial neural network modeling of the relationship between feed properties and H–oil operation data and H–oil VTB properties was also applied.

2. Materials and Methods

2.1. Experimental Materials

Forty-one crude oils, whose properties are summarized in Table S1, were processed in the H–Oil vacuum residue during the period 2015–2025.

The characteristics of the vacuum residues, separated from these 41 crude oils by laboratory TBP distillation following ASTM D 2892 [59] and ASTM D 5236 [60] standards, are presented in

Table 1. Characteristics of straight-run vacuum residues obtained by TBP distillation of the 41 processed crude oils in the studied refinery.

												High Temperature Simulated Distillation (HTSD)
No	Crude oil	SG	MCR, wt.%	Sulfur, wt.%	Vis, °E	Saturate, wt.%	Aromatics, wt.%	Resins, wt.%	C7-asp, wt.%	C5-asp, wt.%	SP, °C	IBP	10%	30%	50%	70%	90%	95%	FBP	Recovery,%
1	Arab Heavy	1.040	23.6	5.8	206	12.4	61.9	wt.5	wt.%	32.9	51.2	524	567	614	653	686			761	85.1
2	Arab Light	1.029	18.7	4.9		15.9	64.7	7.3	12.1	18.8	32.3	498	549	592	629	664	712	734	740	96.6
3	Arab Med.	1.031	20.7	5.4	94.8	11.8	68.3	5.3	14.6	25.5	44.7	513	560	600	633	663			691	84.0
4	Aseng	0.984	14.2	0.6		32.7	48.5	15.2	3.7	10	28	523	556	588	619	656	712	776	776	95.3
5	Azeri Light	0.967	9.5	0.5	17.3	40.2	50.1	8.4	1.4	5.4	30.2	483	526	567	605	644			650	73.1
6	Basrah H	1.071	28.9	7.1	487	12.3	54.1	5.8	27.7	37.0	68.6	488	537	588	626				643	62.6
7	Basrah L	1.052	23.8	5.9	127.5	12.3	64.8	4.9	18	27.7	50.3	507	560	603	637	666	710		713	91.0
8	Basrah Medium	1.057	24.2	6.82	203				22.3	30.2		502	553	598	634	665	711		714	91.8
9	Bonga	0.968	12.8	0.74	35.3	26.4	59	13.9	0.70
10	CPC	0.981	16	2.1	22.5	44.6	40.8	10.3	3.4	11	25.2	487	518	551	584	625			646	79.1
11	El Bouri	1.050	25.5	3.3	139.2	12	57.9	12.6	17.5	27.3	45	478	523	568	610				642	66.2
12	El Sharara	0.976	13.1	0.39	18.3				10.1	17.0		504	542	573	601	637	692	713	756	99.4
13	Es Sider	0.999	13.8	1.096	31				10.2	19.0		505	551	591	628	661	711	733	739	96.1
14	Forties	0.990	14.8	2.5		28.7	60.3	3.8	7.2	9.8	28.9	517	559	596	633	672	738		779	93.4
15	Gulf of Suez	1.024	19.7	3.30	82.2				22.1	32.0		498	556	599	637	671	718			90.2
16	Helm	1.054		3.013	422.3				27.0	41.3		507	554	598	634	665			703	87.1
17	Imported AR_july 2024	1.047	20.8	6.30					19.2	24.9
18	Iranian H	1.050	23.9	5.2		17	52.6	5	25.4	36.2	61.9
19	Johan Sverdrup	1.023	18.26	1.77	138				16.4	27.4		514	557	600	641	679			716	87.3
20	Kazakh H	0.990	17.1	1.7		33	50.2	5.7	11.1	17.8	27.8	410	549	592	632	672	731		771	93.1
21	Kzakh/Kumkol blend	0.990	17.1	1.195	15.75
22	KBT	1.067	26.9	6.4	129.3	12.3	53.6	9.2	24.9	32.4	62.4
23	KEB	1.037	23.3	5.7		15	64.2	4.2	16.6	25.7	47.8	514	560	606	647	682			718	87.1
24	KEBCO	1.020	16.3	3.23	35				14.4	18.9		504	551	591	627	660	708	728	735	97
25	Kirkuk	1.054	25.2	5.9	120.8	15.2	55.4	5	24.3	33.1	58.1	513	558	603	645	680			709	84.7
26	LSCO	0.993	14	1.58	23.8	25	61.1	6.1	7.8	15.5	28.9	508	553	585	592	668	719		730	91.8
27	Okwuibome	0.975	12.9	0.497								509	553	585	616	652	707	760	811	97.9
28	Payara Gold	1.001	13.0	1.43	39.8				8.1	13.2		502	550	591	629	663	710		727	94.7
29	Prinos	1.108	32.8	9.14		12.6	50.6	6.8	30	38.8	69.2	491	539	574	613	649			663	78.8
30	RasGharib	1.059	25.1	5.6		14.7	49.7	9.6	26	34.9	75.8	505	558	606	640	670			675	73.9
31	Rhemoura	1.041	23.7	1.8	42	19.7	49.8	7.3	23.2	31.3	51.1	487	533	577	617				650	67.0
32	Sepia	0.998	13.8	0.75	58				8.5	17.1		510	561	606	641	673	717	740	779	100.6
33	SGC	1.050	22.9	5.09		15	55.9	7.3	21.8	28.4	58.4	490	538	588	627	655			690	88.2
34	Tartaruga	1.008	16.3	1.35	77				14.3	22.4		502	553	597	635	669			708	87.9
35	Tempa rossa	1.120	34.3	9.3		2.2	48.4	12.6	36.8	46.8	100	531	576	627	659	690			696	74.4
36	TEN_Oct.2024	0.981	11.6	1.06	15.8				1.2	5.0		491	544	586	625	663	714	741	817	99.8
37	Unity Gold	0.979	14.7	1.32	36.5				10.9	15.7		503	550	589	626	662	711		729	94.1
38	Urals	0.997	17.5	3	47.5	25.6	52.5	7.8	14.1	17.6	40.1	497	553	595	631	663	710		718	93.3
39	Val’Dagri	1.052	21.4	6	79.5	11.7	73.5	6.4	8.5	19.5	43.7	488	550	592	630	663	707	732	856	107
40	Varandey	0.990	15.1	1.7	24.8	33.5	47.6	11.3	7.6	13.5	43.8	520	559	598	635	674	738		764	92.2
41	Western Desert	1.052	19.0	1.31	60				17.9	24.7		510	547	585	622	663	717		726	92.6

Note: SG = specific gravity at 15.6 °C; MCR = microcarbon residue; Vis = specific viscosity at 120 °C; SP = softening point.

. The properties of the cutter stocks added to the H–Oil VTB in the heavy fuel production process at the studied refinery are presented in

Table 2. Physical and chemical properties of the main cutter stocks used to dilute the H–Oil VTB.

	FCC LCO	FCC HCO	FCC SLO	H–Oil Diesel
Density at 15 °C, g/cm3
Kinematic viscosity at 80 °C, mm2/s	1.42	4.42	33.35	3.56
HTSD, ASTM D-7169, wt.%	0.9399	1.0147	1.1008	0.872
0.5	158	200	247
5	189	257	311
10	200	273	325	201
30	224	306	359
50	245	322	393	269
70	264	339	433
90	292	372	525	330
95	308	393	594
99.5	380	460	701
Sulfur, wt.%	0.2	0.8	1.2	0.2
SARA composition, wt.%
Saturates	19.9	18.2	15.1	45.3
Aromatics	77.1	75.1	50.7	54.7
Resins	0	5.4	27.6	0
Asphaltenes	0	0	3.5	0
Kw-factor	10.4	10.09	9.65	11.37
TSE, %	0	0	0.07	0

.

2.2. Experimental Methods

The methodology applied in this study is summarized in Figure 1.

Figure 2 presents a simplified process diagram of the refinery where the heavy fuel oil was produced during the processing of the 41 crude oils.

Figure 3 depicts the process diagram of the H–Oil vacuum residue hydrocracker where the investigation was carried out.

The methods used to characterize the finished fuel oil, SRVRs, H–Oil VTBs, H–Oil VR feed blend, and partially blended fuel oil (PBFO) are summarized in

Table 3. The methods employed to measure the properties of finished fuel oil, SRVRs, H–Oil VTBs, H–Oil VR feed blend, and partially blended fuel oil (PBFO).

Property	Standard method
Density, kg/m3	BDS EN ISO 3675 [61]
Sulfur content wt.%	ASTM D 4294 [62]
Asphaltene (C7, and C5) content, wt.%	ASTM D 6560 [63]
Microcarbon content, wt.%	EN ISO 10370 [64]
Specific viscosity, °E	ASTM D 1665 [65]
Carbon content, wt.%	ASTM D 5291 [66]
Hydrogen content, wt.%	ASTM D 5291 [66]
Nitrogen content, wt.%	ASTM D 5291 [66]
Nickel, ppm	IP 501 [67]
Vanadium, ppm	IP 501 [67]
Sodium, ppm	IP 501 [67]
Iron, ppm	IP 501 [67]
High-temperature simulation distillation (HTSD)	ASTM D 7169 [68]
Water content, wt.%	ASTM D 95 [69]
Mechanical impurities content, wt.%	ASTM D 473 [70]
Flash point open cup, °C	ASTM D 92 [71]
Flash point closed cup, °C	ASTM D 93 [72]
Pour point, °C	ASTM D 97 [73]
Ash content, wt.%	ASTM D 482 [74]
Specific heat of combustion/lower	ASTM D 4809 [75]
Total sediment existent, wt.%	IP 375 [76]
Total sediment potential, wt.%	IP 390 [77]
Total sediment accelerated, wt.%	IP 390 [77]

.

The assessment of relationships between the characteristics of the finished fuel oil, the straight-run vacuum residues separated from the 41 crude oils, and the H–Oil VTB was performed using intercriteria analysis (ICrA) [78]. Details of the theory and application of ICrA are presented in [79]. The ICrA approach computes two intuitionistic fuzzy functions, μ and υ, whose values define the degree of the relationship between the criteria.

For μ = 0.75 ÷ 1.00 and υ = 0 ÷ 0.25, a region of statistically meaningful positive consonance is determined, while at μ = 0 ÷ 0.25 and υ = 0.75 ÷ 1.00, an area of statistically meaningful negative consonance is derived. All other cases are considered to be dissonance. A strong consonance is considered at values of μ = 0.95 ÷ 1.00, υ = 0.00 ÷ 0.05 (positive) and μ = 0.00÷0.05, υ = 0.95÷1.00 (negative), while a weak consonance is deemed at μ = 0.75 ÷ 0.85, υ = 0.15 ÷ 0.25 (positive) and μ = 0.15 ÷ 0.25, υ = 0.75 ÷ 0.85 (negative). Two software packages for ICrA were established and freely available as open source from https://intercriteria.net/software/ (accessed on 10 April 2025), and they are detailed in [80,81,82].

Before ICrA evaluation, all variables are normalized using the following normalization formula (Equation (1)):

$$X_{new}={\displaystyle \frac{X-X_{min}}{X_{max}-X_{min}}}$$

(1)

Two types of regression models, linear and nonlinear, were employed in this research. The optimal parameters of regression coefficients were established by using the nonlinear least squares method (LSM). For optimization the differential evolution (DE) algorithm was used to search for the best parameter values in compact subsets in parameter space. The DE algorithm does not require differentiability or even continuity of the optimized function. Moreover, it is very applicable to construct some necessary bounds in the parameter space. All calculations have been realized by the use of the CAS Maple 2024.2 Global Optimization Tool (method = diffevol) and verified by a simple Python script based on a differential evolution package from the scipy.optimize library.

The use of artificial neural networks leverages the natural connectivity of individual neurons to create an architecture for predicting the relationship of H–Oil vacuum residue feed blend characteristics and H–Oil operating conditions to the VTB properties. ANN modeling was accomplished using Matlab 2020 software. The training, validation, and testing data sets used by the ANN were split as follows: 70% for training; 20% for testing; and 10% for validation.

3. Results and Discussion

3.1. Investigation of Finished Heavy Fuel Oil Characteristics

Table 4. Ambit of change in properties and component composition of the heavy fuel blend under study.

Characteristics of Finished Heavy Fuel Oil	Min	Max	Average
Density, g/cm3	0.9574	1.0489	1.007
Specific viscosity, °E	4.75	14.97	11.7
Sulfur content wt.%	0.7	2.21	1.3
Water content, wt.%	0.01	0.7	0.1
Mechanical impurities content, wt.%	0.01	0.9	0.1
Flash point open cup, °C	94	212	126.7
Flash point closed cup, °C	97	206	159.6
Pour point, °C	0	21	11.3
Ash content, wt.%	0.011	0.099	0.0
Specific heat of combustion/lower	39.374	41.406	40.3
Total sediment existent, wt.%	0.01	0.47	0.08
Total sediment potential, wt.%	0.01	0.8	0.17
Total sediment accelerated, wt.%	0.02	0.7	0.12
Vanadium content, ppm	25	170	63.3
Content of aluminum and silicon, ppm	25	244	75.7
Microcarbon content, wt.%	8	23.1	14.8
Asphaltene (C7) content, wt.%	2.4	15.3	7.0
Content of components in the heavy fuel oil blend, wt.%
VTB from H–oil	36.9	92.4	71.1
LCO from FCC	0.0	18.8	2.3
HCO from FCC	1.7	62.6	19.8
Slurry oil from FCC	0.0	21.7	5.2
SRAR from CDU	0.0	2.0	0.1
Diesel from H–Oil/CDU	0.0	20.1	0.6
VGO	0.0	5.2	0.2
SRVR from VDU	0.0	13.2	1.7

indicates the range of variation in the characteristics and composition of the components of the 120 heavy fuel samples examined.

Figure 4 visualizes the alteration of main characteristics of the 120 studied finished fuel oil blend samples.

The data in Table 4 show that the fuel oil blend produced based on H–Oil VTB is characterized by a wide range of property fluctuation. The average of the blending recipe shows that the fuel oil mixture contains about 70 wt.% H–Oil VTB and 30 wt.% cutter stocks (diluents). The data in Figure 4a suggest that the microcarbon residue and asphaltene contents in the fuel oil follow the trend of H–Oil conversion. The enhancement of conversion is associated with an increase in MCR and asphaltenes as a result of the concentration of the most refractory poly-nuclear aromatic structures, which have not been cracked to lower molecular weight components boiling in the distillate range [83]. As evident from the data in Table 4, the H–Oil VTB content of fuel oil has varied between 37 and 92%, so the microcarbon residue and asphaltenes content, which originate from the VTB in fuel oil, is not always consistent with the change in conversion. The data in Figure 4b indicate that the sulfur content oscillates around 1.3 ± 0.7 wt.%, while the specific viscosity does not exceed 15 °E at 80 °C. Some peaks in TSP and TSA are observed in the heavy fuel oil samples, with numbers around 60, 90, and 113. In order to evaluate the presence or absence of statistically meaningful relationships between the different properties of the finished heavy fuel oil, ICrA evaluation was performed on the data from the 120 fuel oil samples. Tables S2 and S3 exhibit the μ and υ values of all the investigated properties mentioned in Table 3 for the 120 samples of heavy fuel oil, while

Table 5. μ-values from ICrA evaluation of the data for properties of the 120 studied heavy fuel oil samples showing the presence of statistically meaningful positive and negative consonances.

μ	D15	Ash	Heat	TSE	TSP	TSA	Al + Si	CCR	Asp
D15	1.00	0.65	0.12	0.38	0.48	0.44	0.70	0.88	0.78
Ash	0.65	1.00	0.33	0.46	0.51	0.49	0.75	0.65	0.63
Heat	0.12	0.33	1.00	0.55	0.49	0.52	0.26	0.13	0.20
TSE	0.38	0.46	0.55	1.00	0.76	0.81	0.48	0.37	0.44
TSP	0.48	0.51	0.49	0.76	1.00	0.84	0.56	0.47	0.53
TSA	0.44	0.49	0.52	0.81	0.84	1.00	0.53	0.43	0.49
Al + Si	0.70	0.75	0.26	0.48	0.56	0.53	1.00	0.69	0.67
CCR	0.88	0.65	0.13	0.37	0.47	0.43	0.69	1.00	0.79
Asp	0.78	0.63	0.20	0.44	0.53	0.49	0.67	0.79	1.00

Note: The bold figures indicate statistically meaningful positive or negative consonance.

and

Table 6. υ-values from ICrA evaluation of the data for properties of the 120 studied heavy fuel oil samples showing the presence of statistically meaningful positive and negative consonances.

υ	D15	Ash	Heat	TSE	TSP	TSA	Al + Si	MCR	Asp
D15	0.00	0.32	0.85	0.53	0.47	0.49	0.26	0.09	0.18
Ash	0.32	0.00	0.64	0.45	0.44	0.44	0.25	0.31	0.33
Heat	0.85	0.64	0.00	0.36	0.45	0.41	0.70	0.83	0.77
TSE	0.53	0.45	0.36	0.00	0.13	0.09	0.42	0.54	0.47
TSP	0.47	0.44	0.45	0.13	0.00	0.09	0.38	0.48	0.42
TSA	0.49	0.44	0.41	0.09	0.09	0.00	0.40	0.51	0.45
Al + Si	0.26	0.25	0.70	0.42	0.38	0.40	0.00	0.27	0.29
MCR	0.09	0.31	0.83	0.54	0.48	0.51	0.27	0.00	0.17
Asp	0.18	0.33	0.77	0.47	0.42	0.45	0.29	0.17	0.00

Note: The bold figures indicate statistically meaningful positive or negative consonance.

indicate those fuel oil properties which demonstrate statistically meaningful positive and negative consonances.

The data in Table 5 and Table 6 display the presence of an intermediate positive consonance of fuel oil density with the content of microcarbon residue, an intermediate negative consonance with the heat of combustion, and a weak positive consonance with the asphaltene content. There is a weak positive consonance between the fuel ash content and the contents of alumina and silicon. Weak positive consonances exist between the three methods for measuring the sediment content in the fuel oil (TSE, TSA, and TSP). It is interesting to note that the stability expressed by TSE, TSA, and TSP have no consonance with any other fuel oil characteristics. The asphaltene content, considered to be a major contributor to sediment formation in hydrocracked heavy oils [84], exhibits dissonance with the TSE, TSA, and TSP, confirming again that sediment formation in hydrocracked heavy oils is a rather complex phenomenon [85,86]. Another observation worth discussing is the demonstrated dissonance of H–Oil conversion with the TSE, TSA, and TSP of fuel oil, given that increasing the conversion of vacuum residues leads to an exponential magnification in the hydrocracked heavy oil sediment content [87,88]. One may suggest that this dissonance may result from the use of diverse cutter stocks with different proportions, and that the various vacuum residues may have diverse affinities for hydrocracking sediment formation. Both reasons have been shown to be true in research [30,89,90]. In the following sections, the influence of different straight-run vacuum residues, cutter stocks, and H–Oil operating conditions on the stability of fuel oil will be discussed.

3.2. Contrasting Straight-Run Vacuum Residue Properties Against the Properties of Hydrocracked Vacuum Residues H–Oil VTBs

Wiehe [91], before establishing the oil compatible model, noted that the straight-run vacuum residues never formed asphaltenes precipitates. Indeed, the straight-run vacuum residues explored in this work exhibited S-values, measured according to ASTM 7157 [42], in the range 2.2–4.3 (see Table S4). Alonso et al. [92] provide evidence in their research that all oils which have an S-value > 2.0 are stable. Therefore, the statement by Wiehe was also confirmed in our study. In contrast to the straight-run vacuum residues, the H–Oil VTBs demonstrated S-values between 1.0 and 1.3 for the H–Oil conversion level between 55 and 70 wt.%. The main reason for this very low S-value was the very low solubility of the asphaltenes from H–Oil VTB. By studying the dependence of the solubility of asphaltenes from straight-run vacuum residues and H–Oil VTB it was found that the density of the vacuum residues best correlates with it. Figure 5 shows that the asphaltene solubility in the hydrocracking processes dramatically drops. Furthermore, it shows that as the density of the vacuum residue increases, the solubility of asphaltenes decreases, which is much more noticeable in the case of H–Oil VTB than in the case of SRVR. If the Sa of H–Oil VTBs is extrapolated to the point of the complete insolubility of asphaltenes (Sa of zero) it can be seen that the density of H–Oil VTB at this point would be about t 1.10 g/cm³.

Figure 6 displays that an increase in the H–Oil conversion leads to an exponential enhancement in the H–Oil VTB density and thus in the solubility of asphaltenes.

Along with the increase in density, a decrease in the molecular weight of H–Oil VTB was observed as the vacuum residue conversion increased (Figure 7).

The data in Figure 7 reveal that a decrease of about 150 g/mol in the molecular weight of H–Oil VTB occurs when the conversion goes up from 60 to 90 wt.%. It is understood that not only the molecular weight of the entire H–Oil VTB will decrease with increasing conversion, but also its asphaltene fraction. Buch et al. [93] reported that after residue hydrotreatment the molecular weight of asphaltenes decreased. In general, the lower the molecular weight of asphaltenes, the better their solubility. [94]. Therefore, increasing the H–Oil conversion, on the one hand, worsens the solubility of asphaltenes due to the increase in their density and H/C ratio, and on the other hand, can improve the solubility of asphaltenes due to the decrease in their molecular weight.

Other research that investigated SRVRs and VTBs from H–Oil VTBs and employing the SAR-AD method, following Adams et al. [95], revealed that the chemical composition of these primary and secondary vacuum residues differed significantly. Figure 8a exhibits that while the asphaltene content determined by the SAR-AD method strongly correlates with the total pericondensed aromatics for the SRVRs, for the H–Oil VTBs this correlation does not exist at all (Figure 8b).

The data in Figure 8 suggest that the predominant type of poly-nuclear aromatic structures in the SRVR asphaltenes are of the pericondensed type. This pericondensed type of poly-nuclear aromatic structures, however, appears to change in the hydrocracking environment.

Both investigated SRVRs and H–Oil VTBs demonstrated strong correlations between H content, H/C atomic ratio, microcarbon residue, and density. Figure 9 shows the correlations between these primary and secondary vacuum residues. The relationships between the investigated properties of the straight-run vacuum residues and the hydrocracked vacuum residues (H–Oil VTB) appear linear, while those for the H–Oil blend feed, that consists of vacuum residue, fluid catalytic cracking (FCC) slurry oil (SLO), FCC heavy cycle oil (HCO), and recycled (the gray circles), exhibit a greater dispersion.

It is evident from the data in Figure 9 that both SRVRs and H–Oil VTBs exhibit linear correlations of the properties H content, H/C atomic ratio, microcarbon residue, and density, but the slope for the primary and secondary vacuum residues is different.

Figure 10 demonstrates the much higher viscosity the SRVRs have at the same density than that of the H–Oil VTBs. Moreover, the viscosity scattering at the same density of SRVRs is much larger than that of H–Oil VTBs. Redelius and Soenen [96] deduced that the viscosity of bitumen (vacuum residues) was a result of a combination effect of molecular weights and molecular interactions. Based on solubility parameters determination, they reasoned that the dispersive interactions are the strongest and that aromaticity was important for the strength of the dispersive molecular interactions. Thus, they concluded that the viscosity of a heavy petroleum product might be calculated from molecular weight and aromaticity. Taking into account that density correlates with heavy oil aromaticity [14] and that T_50% correlates with molecular weight [97], it may be supposed that the viscosity of SRVRs and H–Oil VTBs could be estimated from T_50% and density in a manner similar to that described by Kotzakoulakis and George [98], and Sinha et al. [99]. Equations (2) and (3) present the correlations developed by using nonlinear regression techniques to predict the viscosity of SRVRs and H–Oil VTBs. It is worth mentioning here that the H–Oil VTB viscosity was best predicted only from the density with %AAD of 15%, while the prediction of the SRVR viscosity was featured with considerably lower accuracy (%AAD of 45.5%). This can be seen as follows:

$$SRVR VIS={\displaystyle \frac{-2008.71}{{(T_{50\%}+191.6376)}^{0.6595}}}-{\displaystyle \frac{\mathrm{73,809.6351}}{D_{15}^{0.02016}}}+\mathrm{73,885.6118} \mathrm{R}=0.788,\%\mathrm{AAD}=45.5\%$$

(2)

where

T_50% = T_50% of SRVR, °C;

D₁₅ = D₁₅ of SRVR, g/cm³.

$$HOil VTB VIS=2822.22887\times {{VTB D}_{15}}^3+5494.2975-8301.6652\times {VTB D}_{15}$$

$$R=0.968,\%\mathrm{AAD}=15.4\%$$

(3)

where

HOil VTB VIS = specific viscosity of H–Oil VTB at 120 °C, °E;

VTB D₁₅ = density of H–Oil VTB at 15 °C, g/cm³

The results of this research showed that the H–Oil VTB does not need T_50% to adequately model its viscosity, while both density and T_50% turned out to be insufficient parameters to fully describe the molecular interactions in the SRVRs, which affect viscosity. This example confirms again the incomparability of interactions, which take place in the SRVRs and in the H–Oil VTBs.

Figure 10. Variation in vacuum residue viscosity with density fluctuation. The yellow circles are estimated SRVR viscosity by Equation (2). The gray diamonds are calculated H–Oil VTB viscosity by Equation (3).

Figure 10. Variation in vacuum residue viscosity with density fluctuation. The yellow circles are estimated SRVR viscosity by Equation (2). The gray diamonds are calculated H–Oil VTB viscosity by Equation (3).

All measured properties of SRVRs and H–Oil VTBs discussed in this section show that secondary vacuum residues exhibit lower molecular weight, lower viscosity, much lower asphaltene solubility, and different molecular interactions, making the behavior of H–Oil VTBs quite different from that of SRVRs.

3.3. Relation of Operation Conditions, Vacuum Residue Feed Blend Characteristics of the H–Oil Hydrocracker to the VTB and PBFO Properties

The range of variation in properties of the combined feed, VTB, and PBFO, along with the operating conditions for 140 cases studied at the commercial H–Oil hydrocracker, are summarized in

Table 7. Range of variation H–Oil VR feed blend, VTB, ATB, and PBFO properties together with the operating conditions WABT and LHSV for the 140 studied cases.

H–Oil Feed	Min	Max	H–Oil VTB	Min	Max	H–Oil PBFO	Min	Max	Average
D15, g/cm3	0.9255	1.0655	D15, g/cm3	0.983	1.1308	D15, g/cm3	0.9654	1.0829
S, wt.%	1.96	4.6	S, wt.%	0.542	2.14	S, wt.%	0.49	1.85
T50%, °C	469	621	T50%, °C	554	608	Viscosity at 80 °C, mm2/s	41.3	320.5
MW, g/mol	389	694	MW, g/mol	482	655	HCO in PBFO, wt.%	0	54.6	27.6
Nc	28	49	Nc	36	48	LCO in PBFO, wt.%	0	46.5	2.3
N, wt.%	0.21	0.66	N, wt.%	0.36	0.91	SLO in PBFO, wt.%	0	12.6	2.9
H, wt.%	9.37	11.65	H, wt.%	7.7	11.8	VTB in PBFO, wt.%	40	88	67.0
C, wt.%	80.97	89.8	C, wt.%	85.2	91.6
MCR, wt.%	6.2	22.5	MCR, wt.%	15.5	47.2
H/C ratio	1.33	1.65	H/C ratio	1.05	1.58
C7 asp., wt.%	3.40	18.00	C7 asp., wt.%	5.20	28.20
C5 asp., wt.%	4.60	29.50	C5 asp., wt.%	8.80	57.10
V, ppm	59	255	V, ppm	22	265
Ni, ppm	8	84	Ni, ppm	6	91
Na, ppm	8	46	Na, ppm	3	35
Fe, ppm	10	105	Fe, ppm	2	243
FCC SLO in H–Oil feed, wt.%	0.0	20.4	H–oil ATB TSE, wt.%	0.0	0.6	PBFO TSP, wt.%	0.02	3.37
HCO in H–Oil feed, wt.%	0.0	10.2
PBFO Recycle in H–Oil feed, wt.%	0.0	25.1
H–Oil Operating conditions
WABT, °C	405	436
LHSV, h−1	0.10	0.25
Catalyst addition rate, kg/t	0.5	1.6
ΔT-R1/ΔT-R2	0.9	3.1
HDM, %	48.1	96.3
H–Oil Conversion, wt.%	46.5	92.6

.

One can see from the data in Table 7 that all investigated streams exhibit a considerable range of property fluctuation. This is a result of oscillation of vacuum residue blend properties (see the data in

Table 8. Range of property fluctuations of the straight-run vacuum residue blends being processed in the commercial H–Oil hydrocracker during the study.

Property	Min	Max
Density at 15 °C, g/cm3	0.987	1.042
MCR, wt.%	14.6	24.1
Sulfur, wt.%	1.6	5.3
Nitrogen, wt.%	0.3	0.7
Saturates, wt.%	10.3	23.2
Aromatics, wt.%	66.0	77.6
Resins, wt.%	3.8	8.8
C7 asphaltenes, wt.%	3.9	11.9
C5 asphaltenes, wt.%	8.1	20.7

), variations in combined feed composition (contents of FCC SLO, HCO, and recycle changing), and alterations of operating conditions (WABT, LHSV, CAR, ΔT-R1/ΔT-R2).

Comparing the data of H–Oil combined feed (Table 7) with that of straight-run vacuum residue blend (Table 8), it is evident that the SRVR blend exhibits a narrower range of density variation and a broader extent of sulfur content fluctuations. This observation can be attributed to the presence of additional components in the H–Oil combined feed, which are heavier (FCC SLO, PBFO recycle) and have a lower sulfur content (see the data in Table 2).

The H–Oil combined feed properties and the operating conditions were found to have a profound effect on the HDM extent, as indicated in the following Equation (4):

$$\begin{gathered} HDM=-188.031+166.9\times Feed{D}_{15}-3.382\times Feed S-217.529\times LHSV- \\ 0.6592\times Rec.+1.3793\times Conv+11.2812\times CAR+16.4283\times {\displaystyle \frac{∆TR1}{∆TR2}} \mathrm{R}=0.880, \\ \mathrm{st}. \mathrm{error}=5.2\% \end{gathered}$$

(4)

where

Feed D₁₅ = Density of H–Oil combined feed at 15 °C, g/cm³;

Feed S = Sulfur content of H–Oil combined feed, wt.%;

LHSV = liquid hourly space velocity, h⁻¹;

Rec. = Content of recycle of PBFO in H–Oil feed, wt.%;

Conv = H–Oil vacuum residue conversion, wt.%;

CAR = Catalyst addition rate, kg/t;

${\displaystyle \frac{∆TR1}{∆TR2}}$ = Ratio between ΔT in the first reactor to ΔT in the second reactor, °C/C.

The demetallization efficiency, which is proportional to the conversion of asphaltenes [100,101], was reported to affect the sediment content in the hydrocracked residual oils [29]. A higher HDM leads to greater asphaltene conversion [100,101], and consequently a lower sediment content in hydrocracked residual oils [29]. The data in Figure 11 confirm the earlier deduction that higher HDM corresponds to lower residual fuel oil sediment content. On the other hand, it also indicates that colloidal stable residual fuel oil cannot be produced if cutter stocks which are different from the high-aromatic FCC gas oils are used, as explained in [102]. To ensure the production of stable residual fuel oil, the HDM extent should be kept higher than 93%. The observed lower dependence of fuel oil TSP on HDM when it is below 75% may be attributed to the precision limitations of TSP measurement. According to IP-375 [76] and IP-390 [77], the maximum total sediment content for residual fuel oils is defined as 0.5 wt.%. Above this threshold, the precision of TSP measurement is not specified by the standards.

Both the data in Figure 11 and Equation (4) suggest that for vacuum residues with HDM values above 93%, stable residual fuel oil could be manufactured when FCC gas oils are used as cutter stocks. To achieve an HDM greater than 93%, the LHSV, CAR, and a higher proportion of fresh catalyst added in the first reactor (60%) relative to the second reactor (40%) must be optimized based on the feed characteristics. Vacuum residues with higher sulfur content appear to be more difficult to demetallize, as evidenced by the negative term of the regression coefficient associated with the effect of sulfur on the HDM in Equation (4).

In addition to colloidal stability, the sulfur content in the residual fuel oil is a critical characteristic that significantly influences SOx emissions during combustion of the fuel oil. The following equation (Equation (5)) predicting the H–Oil VTB sulfur content was developed by processing the data from the 140 cases studied using multiple linear regression:

$$\begin{gathered} HOil VTB S\left(\%\right)=-1.15408-2.214\times Feed D_{15}+0.47186\times Feed S+ \\ 0.006675\times WABT+4.00644\times LHSV-0.1684\times CAR \mathrm{R}=0.927, \mathrm{st}. \mathrm{error}= \\ 0.14\% \end{gathered}$$

(5)

where

HOil VTB S(%) = Sulfur content in the H–Oil VTB, wt.%;

Conv = H–Oil vacuum residue conversion, wt.%;

WABT = Weight average bed temperature, °C.

Equation (5) indicates that feed characteristics such as density and sulfur content influence the level of sulfur in the H–Oil VTB. An increase in CAR decreases fuel oil sulfur content. To decrease the sulfur level in the H–Oil VTB, the operating conditions WABT and LHSV need to be optimized. The same variables were used to model the H–Oil VTB sulfur level using artificial neural network techniques. Figure 12 illustrates the fitness of the prediction by ANN versus the measured H–Oil VTB sulfur content. The ANN model demonstrates a higher accuracy in predicting the H–Oil VTB sulfur content, with R = 0.964 and standard error of 0.128%.

In neural network modeling, the architecture of the network plays a crucial role in determining predictive performance. In this study, a neural network consisting of six layers was employed to model the H–Oil VTB sulfur content. The structure of the network is 64, 32, 16, 10, 8, and 1 neurons, respectively. That is, the first layer contains 64 neurons, followed by the second with 32, the third 16, the fourth 10, the fifth 8, while the sixth (output) layer contains 1 neuron. This architecture was selected to effectively process the input data of size 5×140. The number of neurons in each successive layer is approximately halved, starting from 64 neurons in the first layer. This configuration is considered optimal for capturing the essential features of the input data. Increasing the number of neurons leads to information redundancy and, consequently, an increase in the mean squared error (MSE), which negatively impacts the modeling performance. In the present case, a mean squared error of 0.012061 was achieved at epoch 5 (as shown in Figure S1).

The density of H–Oil VTB is another important property to be predicted from feed characteristics and operating conditions. Regressing the data allowed us to develop the following equation (Equation (6)):

$$\begin{gathered} Hoil VTB D_{15}=-192.6766+9.05459\times Feed S+2.6256\times Feed C_{7}asp+ \\ 2.8908\times WABT-249.5156\times LHSV \mathrm{R}=0.904, \mathrm{st}. \mathrm{error}=0.013 \mathrm{g}/{\mathrm{cm}}^3 \end{gathered}$$

(6)

where

HOil VTB D₁₅ = Density of H–Oil VTB, g/cm³;

Feed C₇ asp = Content of C₇ asphaltenes in H–Oil combined feed, wt.%;

LHSV = liquid hourly space velocity, h⁻¹.

All of the regression coefficients exhibited a probability value lower than 0.05, indicating that they statistically significantly affect the density of H–Oil VTB. Equation (6) shows that feeds with higher sulfur and asphaltene content yield a heavier H–Oil VTB during hydrocracking. The desired level of density—along with related properties such as the MCR and viscosity of H–Oil VTB—can be achieved through the optimization of feed characteristics and operating conditions.

The modeling of H–Oil VTB density was also conducted using ANN techniques. Figure 13 visualizes the agreement between predicted and measured density values for the studied 140 cases of the H–Oil VTB. The ANN model again demonstrates superior accuracy in predicting this H–Oil VTB property, with an R value of 0.958 and a standard error of 0.009 g/cm³.

The same ANN structure used to predict H–Oil VTB sulfur was employed to model the density of H–Oil VTB at 15 °C. As evident from the data in Figure S2, the mean squared error was 0.00024009 at epoch 6.

Considering that the vacuum residue conversion affects properties of H–Oil VTB such as density (Figure 6), MCR, and viscosity, it was also modeled using multiple linear regression and ANN techniques. The regression Equation (7) indicates that for the 140 cases studied the vacuum residue conversion clearly depends on reaction temperature (WABT) and the reaction time, expressed by the LHSV. Recycling of PBFO in the H–Oil hydrocracker confirms its positive effect on conversion, as already reported in [15,103]. This can be seen in the following:

$$\begin{gathered} HOil conversion \left(wt.\%\right)=-341.674+0.2396\times Rec.+0.9935\times WABT- \\ 38.9524\times LHSV \mathrm{R}=0.959, \mathrm{st}. \mathrm{error}=3.0\% \end{gathered}$$

(7)

Opposite to what was observed during a pilot plant study, where H–Oil feed properties—such as sulfur, MCR, asphaltene, and nitrogen contents—were found to influence the vacuum residue conversion level (as reported in [104]), this research did not observe such an effect. The probable reason for this discrepancy is the higher measurement accuracy of the pilot plant’s conversion data, which reports an uncertainty limit of 1.7% for vacuum residue conversion (standard error of 0.85 wt.%). In contrast, the standard error of Equation (7) is 3.5 times greater than that of the pilot plant. It is understandable that the accuracy of conversion measurement in the commercial H–Oil hydrocracker is expected to be lower than that of the pilot plant. This is due to the much larger number of measurements performed in the commercial unit, each with its own error of measurement, as well as the greater number of interfering factors present in such a system.

Figure 14 presents the prediction of H–Oil conversion by the ANN model. It is evident from these data that the ANN model again shows higher accuracy of prediction compared to that obtained using multiple linear regression, with an R value of 0.9714 and a standard error of 2.2%.

The same structure of the ANN models used to predict H–Oil VTB sulfur and density was also applied to model the H–Oil vacuum residue conversion. As observed from the data in Figure S3, the mean squared error was 0.0014101 at epoch 4.

4. Conclusions

The production of stable very-low-sulfur residue fuel oil is limited by two main factors: (1) incompatibility between the main residue stream and the diluents used; and (2) achieving a very-low-sulfur content in the residue. The ebullated bed vacuum residue hydrocracking technology is capable of providing very-low-sulfur residue (≤0.5 wt.%) when the straight-run vacuum residue has a sulfur content not exceeding 1.8 wt.% and the operating conditions are optimized. It was found that the density of hydrocracked vacuum residue strongly correlates with microcarbon residue content, hydrogen content, H/C atomic ratio, and viscosity. Increasing the vacuum residue hydrocracking conversion via reaction temperature augmentation and LHSV reduction leads to an exponential rise in hydrocracked vacuum residue density and viscosity, along with a decrease in molecular weight from 650 to 500 g/mol when conversion enhances from 50 to 93 wt.%.

This study confirmed that all straight-run vacuum residues are colloidal stable, exhibiting an S-value in the range 2.2–4.3. In contrast, hydrocracked vacuum residues demonstrated an S-value between 1 and 1.38, indicating that they are either incompatible or near incompatible with a very low asphaltene solubility. The hydrocracked vacuum residues display lower molecular weight, lower viscosity, much lower asphaltene solubility, and distinct molecular interactions compared to straight-run vacuum residues, resulting in significantly different behavior.

Regression and artificial neural network (ANN) models were developed to predict the sulfur content, density, conversion level, and HDM extent of hydrocracked vacuum residue. These models demonstrated that all properties depend on operating conditions and feedstock characteristics. All ANN models demonstrated higher accuracy of prediction compared to traditional regression methods.

This study confirmed that both the HDM extent and the cutter stocks used are critical factors controlling the fuel oil stability. Stable fuel oil with total sediment potential lower than 0.1 wt.% can be produced if the HDM is not lower than 93% and only high-aromatic FCC gas oils are employed as viscosity reducers. Straight-run vacuum residues, which have higher density and lower sulfur content, appear to be more favorable for hydrodemetallization.

It has become clear that very-low-sulfur fuel oil with high stability cannot be produced by vacuum residue hydrocracking from all crude oils. Instead, it can only be achieved from crude oils whose sulfur content is not higher than 0.9 wt.% as their corresponding vacuum residues have a sulfur level approximately double that of the crude oil. Additionally, during hydrocracking, the HDM extent of such vacuum residues should not fall below 95% and high-aromatic FCC gas oils must be used as cutter stocks.