Roles of Catalysts and Feedstock in Optimizing the Performance of Heavy Fraction Conversion Processes: Fluid Catalytic Cracking and Ebullated Bed Vacuum Residue Hydrocracking

Abstract

Petroleum refining has been, is still, and is expected to remain in the next decades the main source of energy required to drive transport for mankind. The demand for automotive and aviation fuels has urged refiners to search for ways to extract more light oil products per barrel of crude oil. The heavy oil conversion processes of ebullated bed vacuum residue hydrocracking (EBVRHC) and fluid catalytic cracking (FCC) can assist refiners in their aim to produce more transportation fuels and feeds for petrochemistry from a ton of petroleum. However, a good understanding of the roles of feed quality and catalyst characteristics is needed to optimize the performance of both heavy oil conversion processes. Three knowledge discovery database techniques—intercriteria and regression analyses, and artificial neural networks—were used to evaluate the performance of commercial FCC and EBVRHC in processing 19 different heavy oils. Seven diverse FCC catalysts were assessed using a cascade and parallel fresh catalyst addition system in an EBVRHC unit. It was found that the vacuum residue conversion in the EBVRHC depended on feed reactivity, which, calculated on the basis of pilot plant tests, varied by 16.4%; the content of vacuum residue (VR) in the mixed EBVRHC unit feed (each 10% fluctuation in VR content leads to an alteration in VR conversion of 1.6%); the reaction temperature (a 1 °C deviation in reaction temperature is associated with a 0.8% shift in VR conversion); and the liquid hourly space velocity (0.01 h-1 change of LHSV leads to 0.85% conversion alteration). The vacuum gas oil conversion in the FCC unit was determined to correlate with feed crackability, which, calculated on the basis of pilot plant tests, varied by 8.2%, and the catalyst ΔCoke (each 0.03% ΔCoke increase reduces FCC conversion by 1%), which was unveiled to depend on FCC feed density and equilibrium FCC micro-activity. The developed correlations can be used to optimize the performance of FCC and EBVRHC units by selecting the appropriate feed slate and catalyst.

1. Introduction

The oil refining industry delivers energy to move our vehicles [1]. The transportation of people and goods by airplanes, ships, trains, trucks, buses, and cars is unthinkable for modern mankind without the use of petroleum-refined automotive fuels [2]. Thus, the optimal functioning of this industrial branch is vital to contemporary human society. There are now 825 petroleum refineries operating all over the world [3], and the most profitable refinery units are those that convert materials from the bottom of the barrel into high-value transportation fuels and feeds for petrochemistry [4]. Fluid catalytic cracking (FCC), used in more than 600 units around the world, is the most applicable technology to convert heavy oils (vacuum gas oils and atmospheric residue). Ebullated bed vacuum residue hydrocracking (EBVRHC) is the technology used to hydroprocess over 90% of the world’s vacuum residues, which are hydrocracked [5]. Although there are only 21 EBVRHC units in oil refineries around the world, the conversion level they can achieve may surpass that attained by FCC. Feedstock quality has been determined as the single variable that most affects the performance of the FCC and hydrocracking processes [6,7,8,9,10].

Harding et al. [6], investigating the catalytic cracking of SIHGO (sour imported heavy gas oil) in a laboratory micro-activity test unit (MAT), found out that the hydrocarbon groups making up the SIHGO exhibited the following conversion at the same catalyst-to-oil ration of 5 wt./wt.: 93.8 wt.% for the saturates; 53.6 wt.% for the aromatics; and 8.8 wt.% for the polar aromatic compounds. Navarro et al. [7] explored over one hundred FCC feed samples originating from various geographic locations, including atmospheric and vacuum gas oils, light and heavy coker gas oils, hydrocracked residues, atmospheric residues, and deasphalted oils. They characterized them in terms of simulated distillation, refractive index, Conradson carbon, total and basic nitrogen, sulfur, metal content, hydrogen content, and SARA composition. Then, these FCC feeds were cracked in an advanced catalytic cracking equipment (ACE) unit. The authors [7] established that the conversion at maximum gasoline yield of these FCC feeds varied between 50 and 83 wt.%. They specified that the maximum conversion is a function of specific gravity, refractive index, hydrogen content, and total nitrogen content.

Murray [8] characterized six diverse vacuum residues from all over the world and hydrocracked them in a laboratory ebullated bed hydrocracking unit at the same operating conditions. She observed that the different vacuum residues demonstrated dissimilar conversions, reaching a difference of 10 wt.%. Prajapati et al. and [9] investigated six various vacuum residues with substantial differences in properties (variation in density at 15 °C between 0.9174 and 1.0881 g/cm³; asphaltene content between 0.9 and 28.9 wt.%; and microcarbon residue (MCR) content between 8.6 and 26.9 wt.%) and hydrocracked them using a nano molybdenum catalyst in a laboratory batch reactor at a reaction temperature of 410 °C and reaction pressure of 100 kg/cm² for a total reaction time of 4 h. They found that the different vacuum residues hydrocracked at the conditions mentioned before exhibited variations in conversion level between 70.8 and 83.1%. Their results showed that the vacuum residues with lower contents of asphaltene and MCR were more reactive than those with higher contents of asphaltenes and MCR. They also specified that the vacuum residues that contained higher amounts of materials boiling above 550 °C, resins and asphaltenes were the most difficult to hydrocrack. Pham et al. [10] examined seven dissimilar feedstocks for residue hydrocracking, ranging from vacuum residue solvent deasphalting (SDA) pitch and unconverted oils (UCO) to isolated asphaltenes, and hydrocracked them on a molybdenum octoate oil-soluble dispersed catalyst in a laboratory semi-batch slurry-phase reactor at a reaction temperature of 410 °C and reaction pressure of 110 bars for total reaction times between 0.25 and 12 h. The authors [10] determined that the feedstocks with a higher MCR content had a greater affinity to form coke and a lower tendency to produce de-asphalted oil (DAO). They also found that the colloidal stable feedstocks exhibited a lower tendency to produce coke and a greater inclination to yield DAO and to obtain lower amounts of impurities compared to unstable feedstocks.

Although the catalyst does not affect the conversion level with the same magnitude as the feed quality does in the FCC process, its effect on the overall performance of the FCC unit is substantial [11,12,13]. It controls not only the conversion level but also the selectivity and quality of the FCC products [14,15,16,17]. Over the last three decades, FCC catalysts demonstrated enhancement in activity from 64 to 74–75 wt.% (MAT, or ACE activity) [18,19] as a result of the transition to the short contact FCC unit design, which was aimed to decrease the secondary cracking and to enable operation at a higher severity and processing feeds with deteriorated quality [20,21].

Typically, when a commercial ebullated bed vacuum residue hydrocracker operates at its design capacity, it converts the petroleum vacuum residues to lighter products in the range of 65–70% [22]. This conversion can be further magnified by a proper catalyst type (heterogeneous) selection, along with suitable operating conditions, using a second-type nano-sized catalyst and an adequate fresh catalyst addition system [22,23,24]. The proper catalyst to deal with the plague of the vacuum residue hydrocracking—increased sediment formation at enhanced reaction severity [23,24,25,26,27,28,29,30]—is one that possesses high average pore diameters (a range of macro pore sizes higher than 20 nm). The higher pore diameters turn the catalyst to be more effective in hydrodemetallization (HDM), and hydrodeasphaltization (HDAs), and thus lower sediment formation [31]. The catalysts, which have smaller pore diameters, for example, of approximately 6.5 nm, exhibit very high hydrodesulfurization (HDS) activity and very low activities for HDM and HDAs, and therefore cannot provide good sediment control [31].

Most of the studies dedicated to investigating the effect of feed quality and catalyst characteristics on the performance of FCC and the ebullated bed vacuum residue hydrocracking are performed in laboratory small-scale units. There is a lack of publications that discuss the performance of a commercial FCC and an ebullated bed vacuum residue hydrocracking unit and the effects various feeds and catalysts have on the results of the operation of the both heavy oil conversion processes. For that reason, we have decided to fill this gap by performing an analysis of the operation of a commercial FCC unit and an ebullated bed vacuum residue hydrocracking unit in a complex petroleum refinery configuration while processing feeds with variable quality and using different catalysts. Data mining techniques like intercriteria and regression analyses as well as artificial neural networks are employed in this study to better understand the effects of catalysts and feeds on the performance of both FCC and ebullated bed vacuum residue hydrocracking units. Some laboratory-performed hydrocracking tests are also discussed in this research for better understanding the effect of feedstock quality on the performance of the commercial ebullated bed vacuum residue hydrocracker.

The aim of this study is to quantify the effect of catalyst and feedstock on the performance of the commercial FCC and ebullated bed vacuum residue hydrocracking units.

2. Results and Discussion

2.1. Conversion Variation in a Commercial Vacuum Residue Hydrocracker Investigated by Intercriteria, Regression Analyses, ANN, and Pilot Plant Tests

Considering that the feedstock quality is the single variable that most affects oil refining unit performance [8,9,10,32], one may expect a performance change following the composition of the crude blend processed. Indeed, that was reported with the replacement of the Urals crude oil used by design with other petroleum crudes at the LUKOIL Neftohim Burgas (LNB) refinery (see its processing scheme in Figure S1) in late 2023 and early 2024 [33]. As a result of the lessons learned, the LNB refinery started to select crude oils that are expected to be processed without creating problems, such as increased fouling rate, corrosion rate, and poor performance of the most profitable conversion processes, ebullated bed vacuum residue H-Oil hydrocracking (see its flow diagram in Figure S2) and fluid catalytic cracking (see the FCC process diagram in Figure S3). In the initial process of replacing of Urals crude oil with other crude oils in December 2023, the H-Oil hydrocracker registered a drop in the vacuum residue conversion from 81 down to 72 wt.% with simultaneous increase of sediment content in the atmospheric tower bottom product (ATB). from 0.13 to 0.40 wt.% [33]. This was also associated with an enhanced fouling rate at the bottom of the atmospheric tower. Then, by excluding the identified problematic crude oils, the H-Oil conversion level was enhanced up to 83 wt.% with a reduction in the ATB sediment content down to 0.03 wt.% in July 2024. However, along with the improvement of H-Oil performance, it was observed that the net vacuum residue conversion deviated from the predicted value based on the plug flow reactor model. The parameters of this model established in [34] are as follows: an activation energy of 215 kJ/mol, a reaction order (n) of 1.59, and a collision factor of $k_0=4.42875\times {10}^{15}$. This model was found to predict H-Oil conversion with the highest accuracy in comparison with the CSTR model ($k_0=1.4369\times {10}^{11}$; $E_A=152.67$ kJ/mol; n = 1.82) and the regression model as communicated in [34]. This model had been developed using data from the H-Oil unit while processing vacuum residues derived from crude oil blends of 70% Urals/30% Middle East. Figure 1 indicates a graph of variation of observed H-Oil net conversion and was calculated using the plug flow reactor model.

The discrepancy between the observed net H-Oil conversion and that calculated from the plug flow reactor model may come as a result from the treatment of different feedstocks. In order to verify this hypothesis, laboratory hydrocracking experiments with six distinct vacuum residues were carried out, and a regression model that predicts conversion at the same operating conditions from feed properties was developed. Then, this empirical correlation we called “feed reactivity”, along with commercial H-Oil data, which were evaluated by intercriteria analysis, resulted in a new regression model that predicts H-Oil conversion. Additionally, an artificial neural network (ANN) model was developed to compare its ability to predict H-Oil conversion with the regression model. Both models revealed that the composition of crude oil blends processed in the refinery affects substantially the level of H-Oil conversion.

2.1.1. Determination of the Feed Reactivity in the Ebullated Bed Vacuum Residue Hydrocracking

The conversion in the vacuum residue hydrocracking was found to be thermally driven [35,36], where the catalyst supplies hydrogen to the vacuum residue and prevents its carbonization [36,37], while at the same time removing the impurities (sulfur, nitrogen, and metals) [23,38,39,40]. For that reason, in most cases of vacuum residue hydrocracking, the catalyst has no impact on the conversion level [24], although there are few reports announcing some effect of catalyst type on the conversion [41,42]. Thus, in order to increase the conversion, the vacuum residue hydrocracker operator augments reactor temperature. However, investigations have shown that some components in the vacuum residue can affect the thermal conversion level, for example, sulfur, which increases conversion [43,44,45], and asphaltenes, which retard the vacuum residue conversion [22,46]. In order to find a quantitative relation of the vacuum residue properties with respect to the conversion level obtained at the same operating conditions, a multiple regression analysis of the data reported in [8] was performed in order to define the hydrocracking feed reactivity. The conversion predicted by the multiple regression model developed (Equation (1)) was contrasted against the experimentally measured conversion, and as shown in Figure 2, a good agreement was obtained.

$$\begin{gathered} VR hydrocracking reactivity = f\left(Sulfur, Nitrogen, Asphaltenes, Concarbon \left(SG\right)\right) \\ R = 0.99, \mathrm{standard} \mathrm{deviation} = 1.1 \mathrm{wt}.\% \end{gathered}$$

(1)

where

VR hydrocracking reactivity = vacuum residue conversion, obtained during hydrocracking at the same operating conditions, wt.%;

Sulfur = vacuum residue sulfur content, wt.%;

Nitrogen = vacuum residue nitrogen content, wt.%;

Asphaltenes = vacuum residue C₇ asphaltene content, wt.%;

Concarbon = vacuum residue Conradson carbon content, wt.%;

SG = specific gravity.

The double standard deviation of Equation (1) of 2.2 wt.%, used as an indicator for the uncertainty of prediction [47], is very close to the uncertainty of the laboratory vacuum residue hydrocracking reported by Fortain [48] of 1.7 wt.%.

For confidentiality reasons, the explicit form of Equation (1) cannot be revealed. The essence of Equation (1) reveals that vacuum residues, which have higher sulfur and lower nitrogen contents and are at the same time more aromatic (have a higher density and Conradson carbon content [49]), and also have a lower ratio of asphaltenes to total aromatic carbon fractions (aromatics + resins + asphaltenes), exhibit higher conversion during hydrocracking at the same operating conditions.

Figure 3 shows the difference in the conversion, estimated by Equation (1) relative to the conversion of the base case, which is Urals vacuum residue, for all studied vacuum residues whose properties are shown in

Table 1. Physical and chemical properties of the vacuum residues derived from the explored 17 crude oils and the two imported atmospheric residues used as feeds for the commercial ebullated bed H-Oil hydrocracker.

Vacuum Residues from Processed Crude Oils	Urals	Light Siberian	Tartaruga	Sepia	Johan Sverdrup	KEBCO	CPC	Helm	Basrah Medium	Arab Light	Arab Med	Arab Heavy	Es Sider	Rhemoura	Unity Gold	Gulf of Suez	Payra Gold	Kirkuk AR, Jan. 2024	AR July 2024
Origin	Russia	Russia	Brazil	Brazil	Norway	Kazakhstan	Kazakhstan	Netherlands	Iraq	Saudi Arabia	Saudi Arabia	Saudi Arabia	Lybia	Tunisia	Guyana	Egypt	Guyana	Iraq	Iraq
D 15, g/m3	0.997	0.993	1.008	0.982	1.0225	0.997	0.9477	1.0540	1.0615	1.029	1.031	1.04	0.991	1.041	0.994	1.024	0.9810	1.054	1.047
S, wt.%	3.0	1.58	1.35	0.75	18	3.0	1.4	3.0	6.5	4.9	5.4	5.8	1.05	1.8	1.318	3.4	1.6	5.9	6.3
N, wt.%	0.5	0.42	0.922	0.916	0.697	0.5	0.311	0.344	0.317	0.28	0.357	0.437	0.731	0.500	0.52	0.38	0.727	0.518	0.518
CCR, wt.%	17.5	14	16.3	13.82	19.9	17.5	8.38	23.25	27.05	18.7	20.7	23.6	15.6	23.7	14.7	19.7	13.05	25.2	20.8
Sat., wt.%	19.3	20.6	16.2	24.4	12.7	19.3	40.6	7.3	6.3	11.4	11.0	9.4	21.3	9.2	20.3	12.5	24.8	12.3	13.7
Aro, wt.%	69.1	72.5	74.5	68.5	75.9	69.3	47.1	75.6	77.4	82.6	76.2	74.8	68.6	73.4	73.8	71.3	65.5	75.7	71.4
Res., wt.%	6.59	3.88	3.4	3.67	4.6	6.4	8.45	5.9	7.1	3.03	6.7	7	5.38	7.8	2.2	7.0	6.3	8.0	6.9
C7 asp, wt.%	5.01	3.02	5.9	3.4	6.8	5	3.85	11.2	9.2	3.0	6.1	8.8	4.7	9.6	3.7	9.2	3.4	4.0	8.0
C5 asp, wt.%	11.6	6.9	9.3	7.1	11.4	11.4	12.3	17.1	16.3	6.0	12.8	15.8	10.1	17.4	5.9	16.2	9.7	12.0	15
Na, mg/kg	16.2	7.9	75.3	20.3	< 1.0	9.8	14.2	30.8	6.9	1.5	38.0	3.4	55.1	57.7	8.6	7.0	3.6	71.9	19.2
Ni, mg/kg	10.8	18.2	10.0	19.2	32.5	26.9	12.3	56.1	22.4	17.0	31.8	20.6	35.5	77.5	15.2	43.3	22.2	5.2	21.9
V, mg/kg	89.8	33.2	20.3	21.5	45.4	134.5	84.9	177.0	114.3	35.0	43.5	82.0	12.0	57.0	45.1	123.9	59.3	17.8	86.9
Fe, mg/kg	11.4	10.9	35.3	20.9	59.5	24.4	20.6	21.1	17.7	3.6	17.5	8.8	6.8	26.6	19	20.3	14.9	6.1	8.7
As, ppb	21	55.3	< 10	123.8	36	178	26.1	24	5.6	33.8	21	84.2	29.7	18	52.6	31.6	21.8	28.4	51.0
Ca, mg/kg	5.6	20.5	13.5	3.9	8.5	4.2	7.4	8.4	2.5	1.0	63.1	2.5	18.6	90.5	7.3	5.5	1.0	4.6	6.7

. These vacuum residues have been hydrocracked in the commercial H-Oil hydrocracker in the LNB refinery. The Urals vacuum residue has been selected as a base case because the LNB H-Oil hydrocracker was designed to process this vacuum residue. The data in Figure 3 exhibits the presence of significant variation (up to 16.4 wt.% difference in the estimated conversion) in the reactivity of the processed vacuum residual oils in the LNB H-Oil hydrocracker within the scope of this study. This reactivity difference could be the reason for the observed deviation between observed and estimated conversion with the activation energy of 215 kJ/mol and the reaction order of 1.59, and the collision factor of $k_0=4.42875\times {10}^{15}$ established for the vacuum residue blend 75% Urals/25% Light Siberian [34] (see Figure 1).

2.1.2. Ebullated Bed Vacuum Residue Commercial and Pilot Plant Results at Different Operating Conditions

In order to search for relations between the operating conditions (reaction temperature (WABT), and reaction time $\tau ={\displaystyle \frac{1}{LHSV}}$), conversion, yields, sediment content in hydrocracked residual oils, and mixed vacuum residue feed quality, an ICrA evaluation of data of 185 days of LNB commercial H-Oil unit (for the period 1 January 2024–10 July 2024) was performed. The range of variation of the LNB commercial H-Oil unit variables being analyzed by ICrA is given in

Table 2. Range of variation of variables from the commercial ebullated bed vacuum residue hydrocracker under study.

Parameter	Min	Max
LHSV, h−1	0.133	0.175
Gas yield, wt.%	5.9	8.8
Naphtha yield, %	4.9	9.9
Diesel yield, %	25.0	38.4
VGO yield, %	29.7	43.6
VTB yield, %	13.4	24.6
Balance Conversion, wt.%	72.4	84.1
PBFO TSP, wt.%	0.02	0.59
ATB TSE, wt.%	0.01	0.25
Net conversion, %	71.5	84.6
HCO in H-Oil feed, wt.%	0.00	12.10
Slurry oil in H-Oil feed, wt.%	8.4	20.2
%VR in H-Oil feed	60.0	90.4
WABT, °C	423	433
Estimated conversion with plug flow reactor modelTANGwith EA = 215 kJ/mol; n = 1.59, and $k_0=4.42875\times {10}^{15}$ as described in [6] (called kinetic conversion)	73.6	86.7
VR feed Conradson carbon content, wt.%	14.7	23.6
VR feed sulfur content, wt.%	1.9	5.2
VR feed nitrogen content, wt.%	0.34	0.64
VR feed saturate content, wt.%	11.3	22.8
VR feed aromatic content, wt.%	61.6	77.0
VR feed resin content, wt.%	2.97	9.55
VR feed C7 asphaltene content, wt.%	4.3	14.7
VR feed C5 asphaltene content, wt.%	8.7	24.0
VR feed Na content, ppm	6.0	31.3
“Feed reactivity” calculated by Equation (1), wt.%	Base − 4.9	Base + 8.2
Fresh catalyst addition rate, kg/t feed	0.00	2.52

.

The dataset of 185 days of operation of the H-Oil vacuum residue hydrocracker was assessed by using ICrA.

Table 3. μ—values obtained from ICrA evaluation of the dataset of 185 days of operation of H-Oil vacuum residue hydrocracker.

μ	LHSV	Gas, %	Naphtha, %	Diesel, %	VGO, %	VTB, %	PBFO TSP	ATB TSE	Net conv	HCO%	FCC SLO	%VR in H-Oil feed	WABT	Kin. Conv.	Feed Reactivity	CAR
LHSV	1.00	0.31	0.30	0.37	0.64	0.56	0.48	0.51	0.31	0.45	0.45	0.38	0.35	0.33	0.41	0.42
Gas, %	0.31	1.00	0.70	0.81	0.20	0.29	0.45	0.31	0.70	0.67	0.51	0.43	0.69	0.81	0.47	0.32
Naphtha, %	0.30	0.70	1.00	0.66	0.24	0.34	0.44	0.40	0.71	0.55	0.41	0.57	0.56	0.68	0.50	0.34
Diesel, %	0.37	0.81	0.66	1.00	0.21	0.24	0.51	0.30	0.67	0.77	0.49	0.33	0.73	0.83	0.47	0.28
VGO, %	0.64	0.20	0.24	0.21	1.00	0.56	0.45	0.50	0.38	0.32	0.53	0.48	0.14	0.19	0.56	0.59
VTB, %	0.56	0.29	0.34	0.24	0.56	1.00	0.41	0.58	0.21	0.26	0.53	0.60	0.23	0.29	0.40	0.55
PBFO TSP	0.48	0.45	0.44	0.51	0.45	0.41	1.00	0.47	0.42	0.57	0.50	0.35	0.42	0.48	0.40	0.38
ATB TSE	0.51	0.31	0.40	0.30	0.50	0.58	0.47	1.00	0.33	0.35	0.45	0.48	0.29	0.33	0.33	0.47
Net conv.	0.31	0.70	0.71	0.67	0.38	0.21	0.42	0.33	1.00	0.55	0.44	0.56	0.52	0.64	0.61	0.37
HCO%	0.45	0.67	0.55	0.77	0.32	0.26	0.57	0.35	0.55	1.00	0.51	0.23	0.61	0.67	0.41	0.34
FCC SLO	0.45	0.51	0.41	0.49	0.53	0.53	0.50	0.45	0.44	0.51	1.00	0.40	0.39	0.46	0.44	0.53
%VR in H-Oil feed	0.38	0.43	0.57	0.33	0.48	0.60	0.35	0.48	0.56	0.23	0.40	1.00	0.29	0.38	0.57	0.51
WABT	0.35	0.69	0.56	0.73	0.14	0.23	0.42	0.29	0.52	0.61	0.39	0.29	1.00	0.80	0.27	0.21
Kin. Conv.	0.33	0.81	0.68	0.83	0.19	0.29	0.48	0.33	0.64	0.67	0.46	0.38	0.80	1.00	0.40	0.27
Feed reactivity	0.41	0.47	0.50	0.47	0.56	0.40	0.40	0.33	0.61	0.41	0.44	0.57	0.27	0.40	1.00	0.47
Catalyst addition rate, kg/t feed (CAR)	0.42	0.32	0.34	0.28	0.59	0.55	0.38	0.47	0.37	0.34	0.53	0.51	0.21	0.27	0.47	1.00

and

Table 4. υ—values obtained from ICrA evaluation of the dataset of 185 days of operation of H-Oil vacuum residue hydrocracker.

υ	LHSV	Gas, %	Naphtha, %	Diesel, %	VGO, %	VTB, %	PBFO TSP	ATB TSE	Net conv.	HCO%	FCC SLO	%VR in H-Oil Feed	WABT	Kin. Conv.	Feed Reactivity	CAR
LHSV	0.00	0.65	0.66	0.59	0.33	0.40	0.43	0.36	0.66	0.49	0.52	0.59	0.45	0.64	0.55	0.47
Gas, %	0.65	0.00	0.28	0.16	0.78	0.68	0.46	0.56	0.28	0.27	0.46	0.54	0.10	0.15	0.50	0.56
Naphtha, %	0.66	0.28	0.00	0.31	0.73	0.63	0.47	0.47	0.27	0.39	0.56	0.40	0.23	0.28	0.48	0.54
Diesel, %	0.59	0.16	0.31	0.00	0.77	0.74	0.40	0.58	0.31	0.17	0.48	0.64	0.07	0.13	0.51	0.60
VGO, %	0.33	0.78	0.73	0.77	0.00	0.42	0.46	0.37	0.60	0.62	0.44	0.49	0.66	0.77	0.41	0.29
VTB, %	0.40	0.68	0.63	0.74	0.42	0.00	0.49	0.30	0.77	0.67	0.44	0.37	0.57	0.67	0.58	0.33
PBFO TSP	0.43	0.46	0.47	0.40	0.46	0.49	0.00	0.36	0.49	0.32	0.41	0.56	0.33	0.42	0.51	0.45
ATB TSE	0.36	0.56	0.47	0.58	0.37	0.30	0.36	0.00	0.55	0.50	0.42	0.39	0.45	0.53	0.55	0.34
Net conv.	0.66	0.28	0.27	0.31	0.60	0.77	0.49	0.55	0.00	0.39	0.53	0.42	0.27	0.32	0.36	0.51
HCO%	0.49	0.27	0.39	0.17	0.62	0.67	0.32	0.50	0.39	0.00	0.43	0.71	0.17	0.26	0.53	0.51
FCC SLO	0.52	0.46	0.56	0.48	0.44	0.44	0.41	0.42	0.53	0.43	0.00	0.57	0.41	0.50	0.53	0.35
%VR in H-Oil feed	0.59	0.54	0.40	0.64	0.49	0.37	0.56	0.39	0.42	0.71	0.57	0.00	0.51	0.58	0.41	0.37
WABT	0.45	0.10	0.23	0.07	0.66	0.57	0.33	0.45	0.27	0.17	0.41	0.51	0.00	0.01	0.52	0.55
Kin. Conv.	0.64	0.15	0.28	0.13	0.77	0.67	0.42	0.53	0.32	0.26	0.50	0.58	0.01	0.00	0.56	0.61
Feed reactivity (Equation (1))	0.55	0.50	0.48	0.51	0.41	0.58	0.51	0.55	0.36	0.53	0.53	0.41	0.52	0.56	0.00	0.41
Catalyst addition rate, kg/t feed (CAR)	0.47	0.56	0.54	0.60	0.29	0.33	0.45	0.34	0.51	0.51	0.35	0.37	0.55	0.61	0.41	0.00

present the μ and υ values of the ICrA evaluation of this dataset. The data in Table 3 and Table 4 indicate that the yields of gas, naphtha, and vacuum tower bottom (VTB) products have statistically meaningful consonances with the net conversion. The variables, which were expected to have appreciable influence on the net conversion, like the weight average bet temperature (WABT), the liquid hourly space velocity (LHSV), and the feed reactivity, were not identified by ICrA as statistically affecting the H-Oil conversion for the investigated dataset of 185 days of operation of the H-Oil vacuum residue hydrocracker. Neither did the sediment content in the hydrocracked atmospheric residue (ATB TSE) exhibit any statistically significant consonance with all variables examined in Table 3 and Table 4. However, when the laboratory hydrocracking experiments shown in Figure 4 as a graph of the dependence of conversion, product yields, and hydrocracked atmospheric residue sediment content on reactor temperature are assessed by ICrA (see

Table 5. μ—values obtained from ICrA evaluation of laboratory hydrocracking plant.

μ	WABT	Conversion	ATB TSE	Gas Yield	Naphtha Yield	Diesel Yield	VGO Yield	VTB Yield
WABT	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00
Conversion	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00
Gas yield	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00
Naphtha yield	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00
Diesel yield	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00
VGO yield	0.00	0.00	0.00	0.00	0.00	0.00	1.00	1.00
VTB yield	0.00	0.00	0.00	0.00	0.00	0.00	1.00	1.00

and

Table 6. υ—values obtained from ICrA evaluation of laboratory hydrocracking plant.

υ	WABT	Conversion	ATB TSE	Gas Yield	Naphtha Yield	Diesel Yield	VGO Yield	VTB Yield
WABT	0.00	0.00	0.00	0.00	0.00	0.00	1.00	1.00
Conversion	0.00	0.00	0.00	0.00	0.00	0.00	1.00	1.00
ATB TSE	0.00	0.00	0.00	0.00	0.00	0.00	1.00	1.00
Gas yield	0.00	0.00	0.00	0.00	0.00	0.00	1.00	1.00
Naphtha yield	0.00	0.00	0.00	0.00	0.00	0.00	1.00	1.00
Diesel yield	0.00	0.00	0.00	0.00	0.00	0.00	1.00	1.00
VGO yield	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00
VTB yield	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00

), one can see that the WABT has a very strong influence on the vacuum residue conversion (μ—value of 1.00 and υ—value of 0.00).

The same is evident from the data in Table 5 and Table 6, which show the relation of reactor temperature (WABT) against the hydrocracked atmospheric residue sediment content (ATB TSE) (μ—value of 1.00, and υ—value of 0.00), while in the commercial H-Oil unit the relation between WABT, and ATB TSE (μ and υ values are 0.29 and 0.45, respectively) evaluated by ICrA (see Table 3 and Table 4) suggests no effect of reactor temperature on the sediment formation rate exists. Such a discrepancy between the performance observed in the laboratory and the commercial hydrocracking units may be ascribed to the many variables acting simultaneously in the commercial unit, thus generating a strong noise, contrary to the laboratory unit, where the noise is much lower.

The employment of a regression analysis of the dataset of 185 days of operation of the H-Oil vacuum residue hydrocracker to assess the effect of reactor temperature, LHSV, H-Oil feed vacuum residue content, and feed reactivity (calculated by Equation (1)) indicated that these variables had statistically meaningful effect on the vacuum residue net conversion. The developed regression Equation (2), shown below, demonstrates a multiple correlation coefficient of 0.808 which could be considered statistically significant (R ≥ 0.75 implies the presence of a statistically meaningful relation).

$$\begin{gathered} H-Oil conv.=0.16068\times H-Oil feed VR\mathrm{\%}+0.5665\times Feed Reactivity+ \\ 0.79293\times WABT-84.582\times LHSV-299.7` \\ R = 0.808, \mathrm{standard} \mathrm{deviation} \mathrm{of} 1.96 \mathrm{wt}.\% \end{gathered}$$

(2)

where

H-Oil conv. = net conversion of vacuum residue in the commercial H-Oil hydrocracker, wt.%;

H-Oil feed VR% = content of vacuum residue (the material boiling above 540 °C) in H-Oil feed, wt.%;

Feed Reactivity = calculated by Equation (1) vacuum residue conversion obtained at the same operating conditions in the laboratory hydrocracking plant, wt.%;

WABT = weight average bed temperature of both reactors in the commercial H-Oil vacuum residue hydrocracker, °C;

LHSV = liquid hourly space velocity, h⁻¹.

The p-values of all regression coefficients of Equation (2) are much lower than the significance level of 0.05, and therefore, they can be reckoned to have a statistically meaningful effect on the vacuum residue conversion. The results of ICrA and regression analysis evaluation of the commercial H-Oil vacuum residue hydrocracker data indicate that while ICrA detects the presence or absence of statistically meaningful relations of the individual variables, the regression analysis can discover the presence of a statistically meaningful relation of a combination of variables to the target variable, which individually have been found to have statistically meaningless relation to the target variable. The meaning of Equation (2) is that the H-Oil conversion depends not only on the operating variables WABT and LHSV but also on the reactivity of the vacuum residue mixture calculated by Equation (1) and the content of vacuum residue in the H-Oil mixed feed that can consist of vacuum residue, FCC slurry oil, FCC HCO, and vacuum gas oil.

The same variables, along with a more detailed characterization of the vacuum residue blend, including the contents of Conradson carbon, sulfur, nitrogen, saturates, aromatics, resins, C₅-asphaltenes, and C₇-asphaltenes, were used as input variables in an ANN model. Figure S1, Figure 5 and Figure 6 present the results of ANN modeling of the commercial H-Oil vacuum residue conversion. Comparing the data in Figure 7 with the reported multiple correlation coefficient of Equation (2) (R = 0.808), one can see the superiority of the ANN model (R = 0.942 for the test set) that predicts the H-Oil commercial hydrocracker vacuum residue conversion with a higher accuracy than the regression model (Equation (2)).

Figure S5 illustrates how the neural network training process has been accomplished to build the ANN commercial H-Oil hydrocracker vacuum residue conversion prediction model. It is evident from the data in Figure S5 that the maximum number of epochs allocated for training was 5000, the time for the ANN training took 12 s, the desired mean squared error was set at $1\times {10}^{-6}$, as well as the achieved mean squared error was at $1.49\times {10}^{-5}$, and the training method used was that of Levenberg–Marquardt.

Figure 5 indicates graphs of the neural network training performance for prediction of commercial H-Oil hydrocracker vacuum residue net conversion. One can see from the data in Figure 5 the three types of data: training, testing, and validation. The graph shows that the achieved mean squared error was 0.00028062, which occurred at epoch 5. After epoch 5, the mean squared error began to magnify for both testing and validation, meaning that no further improvement in the model accuracy could be expected.

Figure 6 is a cross-plot of predicted versus measured (target) H-Oil hydrocracker vacuum residue net conversion for the three types of data: training, testing, and validation. The correlation coefficients of the three types of data were found to be R = 0.9623 for training, R = 0.94151 for testing, R = 0.87062 for validation, and overall R = 0.946. The overall data comprised all subsets of data for training (70% of all data), testing (20% off all data), and validation (10% of all data).

2.2. Performance of Commercial Ebullated Bed Vacuum Residue H-Oil Hydrocracker during Employment of Cascade and Parallel Mode of Fresh Catalyst Addition

The effect of catalyst condition on the performance of commercial ebullated bed vacuum residue H-Oil hydrocracking during the employment of cascade mode of fresh catalyst addition is exemplified in the data shown in Figure 7 and Figure 8. They exhibit how the sediment content in the hydrocracked atmospheric residue suddenly went up at the beginning of the second (2018–2021) and third cycles (2021–2025) of the LNB H-Oil hydrocracker. The hydrocracked atmospheric residue sediment content at the start of the second cycle after 20 days on stream commenced enhancing and went beyond the maximum acceptable level of 0.3 wt.% (Figure 7). On the 51st day of the second cycle, the sediment content reached the excessively high value of 1.1 wt.%.

Figure 8 shows exactly the same pattern of dynamics of the atmospheric residue hydrocracking sediment content change over time recorded at the beginning of the third cycle (2021–2025), when the fresh catalyst cascade addition system continues to be used as shown in Figure 7. After 30 days from the start of the third cycle, the sediment level settled in increments beyond the maximum acceptable level of 0.3 wt.% and reached 0.96 wt.% on the 48th day. In both cycles, the increased sediment formation rate required a decrease in the reaction temperature, which ranged from 419 to 410 °C in the second cycle (Figure 7) and from 428 to 415 °C in the third cycle (Figure 8). It is interesting to note that regardless of using three different Ni–Mo-supported solid catalysts, they all exhibited the same poor sediment control at the beginning of the second and third cycles. The common in both cases was the formation of black powder in the first reactor that substituted the solid catalyst in this reactor and represented 80% of reactor inventory. Figure 9 shows photographs of the catalyst unloaded from the first ebullated bed reactor (the left-hand side picture) and from the second ebullated bed reactor (the right-hand side picture) during 2018, and 2021 maintenance works. During the turnaround in Spring 2018 and Autumn 2021, the first reactor was loaded with fresh catalyst only, while the second reactor remained with the existing catalyst from the previous cycle.

The analysis of black powder revealed that it consists of metal sulfide particles as FeV2S4, NiV2S4, and (Ni, Fe)S and no presence of the original Ni–Mo-supported catalyst. It appeared that the catalyst inventory of the first reactor (R-1001) has been progressively replaced during the normal operation of the H-Oil unit by this material. This replacement has obviously occurred over a long period of time considering the concerned quantity. Moreover, the specific size and density of these particles have allowed them to co-exist with the equilibrium catalyst in the ebullated process. The analytical results for the size of the spent catalyst withdrawn from the first reactor over the same period showed it could not be attributed to an unusual breakage of the catalyst inside the reactor. Figure 10 displays the results of fluorescence wave dispersion X-ray (WDXRF) spectrometry of the content of V, Ni, and Fe in three spent catalyst samples taken from the bottom of the first reactor, which represented 20% of reactor inventory, and three samples of black powder extracted from the first reactor, which was 80% of first reactor inventory during the maintenance work in Autumn 2021.

It is evident from these data that the contents of V, Ni, and Fe in the black powder are 3 times, 1.5 times, and 10 times as high as those of the spent catalyst, respectively. Figure 11 demonstrates that the ratios ${\displaystyle \frac{V}{Ni}}$ and ${\displaystyle \frac{V}{Fe}}$ in the black powder are the same as those of the H-Oil feed and much different from those in the spent catalyst, suggesting that the origin of the black powder comes from the feed.

Scanning electron microscopy (SEM) (10 and 20 µm) and microprobe analysis of catalyst samples taken from the first and second reactors during the third cycle unveiled the following:

• The spent catalyst sampled from the first reactor indicated that the catalyst wall is quite homogeneous (Figure 12a), with a high concentration of Fe (15%) and V (34%) at the periphery. The content of iron in the whole catalyst particle from the first reactor was 0.3 wt.%.
• The spent catalyst sampled from the second reactor exhibited that the catalyst wall is not homogeneous (Figure 12b), with a higher concentration of Fe (39%) and V (54%) than the catalyst from the first reactor at the periphery. The content of iron in the whole catalyst particle from the second reactor was between 0.6 and 1.0 wt.%. A thin layer adhered to the outer catalyst surface with a thickness of between 7 and 15 μm was observed (see Figure 13).
• The thin layer adhered to the outer surface of the second reactor spent catalyst was found to contain the same metal sulfides as those identified in the black powder.

The cascade fresh addition system distinguishes with fresh catalyst addition in the second reactor, withdrawal of spent catalyst from the bottom of the second reactor and addition of the second reactor spent catalyst in the first reactor, and withdrawal from the bottom of the first reactor the spent catalyst from this reactor (see Figure S6). The spent catalyst withdrawn from the second reactor is cooled from 420–430 °C down to 232 °C by the transport oil (heavy vacuum gas oil). Then, it is heated up again to 420–430 °C when it enters the first reactor. The catalyst undergoes a thermal shock, and the thin, not homogeneous layer of metal sulfides forms on the catalyst surface in the second reactor, which seems to be broken in the first reactor, releasing these metal sulfides while forming this way a black powder. All these findings suggest that the black powder originates from the not homogeneous metal-containing layer, formed on the outer surface of the catalyst in the second reactor. Considering that the parallel fresh catalyst addition system independently introduces catalyst in both reactors without using the spent catalyst from the second reactor as a catalyst for the first reactor, this system was chosen instead of the cascade one. Moreover, the parallel system enables the control of the activity of the catalytic system in both reactors independently, which, as detailed in [50], allows better sediment control. Now the LNB H-Oil hydrocracker has been employing the parallel fresh catalyst addition system for two years.

The formation of black powder and the related loading of fresh catalyst in the first reactor during the maintenance works in Spring 2018 and Autumn 2021 were associated with poor sediment control and were far from the optimal performance of the commercial H-Oil hydrocracker, and restoring optimal sediment control took 130 days in both the second and third cycles (Figure 7 and Figure 8).

2.3. Conversion Variation in a Commercial Fluid Catalytic Cracking Investigated by Intercriteria, Regression Analyses, and ANN

Seventeen crude oils and two imported atmospheric residues were processed in the LNB refinery under study. The properties of these crude oils are presented in Table S1. The vacuum gas oils derived from these 19 refinery feed oils were cracked in the LNB FCC unit, and their properties are summarized in

Table 7. Characteristics of the vacuum gas oils obtained from the explored 17 crude oils and the two imported atmospheric residues used as feeds for the commercial FCC unit.

Vacuum Gas Oils from Processed Crude Oils	Urals	Light Siberian	Tartaruga	Sepia	Johan Sverdrup	KEBCO	CPC	Helm	Basrah Medium	Arab Light	Arab Med	Arab Heavy	Es Sider	Rhemoura	Unity Gold	Gulf of Suez	Payra Gold	AR, Jan. 2024	AR July 2024
Origin	Russia	Russia	Brazil	Brazil	Norway	Kazakhstan	Kazakhstan	Netherlands	Iraq	Saudi Arabia	Saudi Arabia	Saudi Arabia	Lybia	Tunisia	Guyana	Egypt	Guyana	Iraq	Iraq
Density at 15 °C, g/m3	0.917	0.907	0.930	0.921	0.923	0.919	0.889	0.940	0.933	0.923	0.924	0.930	0.897	0.905	0.909	0.886	0.925	0.929	0.929
Sulfur, wt.%	2.002	0.825	0.799	0.409	0.950	2.112	1.025	1.462	3.701	2.543	2.943	3.334	0.494	0.740	0.573	1.738	0.776	3.164	3.164
Nitrogen, wt.%	0.150	0.112	0.292	0.263	0.181	0.150	0.069	0.168	0.100	0.075	0.077	0.080	0.148	0.112	0.150	0.050	0.050	0.131	0.131

. Availing the correlations developed by Navarro et al. [7] to predict FCC conversion at maximum gasoline yield (so-called crackability) from feed hydrogen content and that of Stratiev et al. [51] relating FCC feed density to the hydrogen content, the following expression was established:

$$Crackability={\displaystyle \frac{85.87257}{1+399.71\times e^{-0.6393\times (33.37-23.12\times d_{15})}}}$$

(3)

where

Crackability = FCC conversion at maximum gasoline yield, observed in laboratory ACE FCC unit wt.%;

d₁₅ = density at 15 °C of FCC feed, g/cm³.

By using Equation (3) and the data from Table 7, the crackability of the 19 vacuum gas oils converted in the commercial FCC unit was obtained and shown in Figure 14.

The data in Figure 14 display that the various vacuum gas oils (VGOs) have quite a different crackability, ranging from 69.5 to 77.7 wt.%. These vacuum gas oils, however, were cracked in the commercial FCC unit as blends and not individual VGOs. They were cracked on seven diverse FCC catalysts, whose characteristics are epitomized in

Table 8. Features of the catalysts employed in the commercial FCC unit during the study.

Catalyst Designation	A	B	C	D	E	F	G
Fresh catalyst properties
Total surface area, m2/g	287	310	325	280	326	298	283
Na2O, %	0.28	0.25	0.3	0.26	0.32	0.29	0.25
RE2O3, %	1.3	1.6	1.9	2.7	1.6	2.8	3.1
Al2O3, %	43.8	40	40.9	42.5	41	42	42
APS, μ	79	75	78	80	79	86	76
ABD, g/cm3	0.72	0.77	0.69	0.71	0.69	0.72	0.71
Pore volume, cm3/g	0.42		0.45	0.40	0.43	0.41	0.43
Micro-activity, %	79	77	78	81	77	81	82
Equilibrium catalyst properties
Total surface area, m2/g	139	160	173	146	164	156	155
Na2O, %	0.15	0.14	0.22	0.19	0.19	0.18	0.18
RE2O3, %	1.33	1.61	1.86	2.7	1.71	2.65	3.1
Al2O3, %	43.6	42.8	43.2	44.3	44.8	45.2	43.1
APS, μ	91	86	95	90	95	104	84
ABD, g/cm3	0.89	0.91	0.82	0.88	0.83	0.85	0.85
Unit cell size, Å	24.28	24.29	24.27	24.33	24.27	24.32	24.34
Micro-activity, %	73.7	71.8	72.6	75	71.1	72.4	75.5
Coke factor	1.02	0.92	0.8	0.85	1.06	0.96	0.87
Ni, ppm	67	108	81	46	109	87	50
V, ppm	169	295	212	159	162	216	155

.

Table 9. Range of variation of variables from the commercial FCC unit and the H-Oil unit, which affect FCC unit performance.

Parameter	min	max
E-Cat activity, %	69.0	77.1
FCC unit feed rate, t/h	78.0	253.0
%H-Oil VGO in FCC feed	0.0	100.0
FCC conversion, wt.%	48.2	82.5
H-Oil WABT, °C	353.0	433.0
H-Oil feed rate, t/h	129.2	310.0
H-Oil Feed FCC SLO,%	3.4	19.2
Riser outlet temperature (ROT), °C	526.0	550.0
Combined feed temperature (CFT), °C	253.3	341.2
Regenerator dense phase temperature (TRG den.) °C	672.0	718.4
Regenerator dilute phase temperature (TRG dil.) °C	682.0	725.3
Hydrogen in coke,%	4.5	7.2
Coke yield, wt.%	4.2	6.5
Catalyst-to-oil ratio (CTO), wt./wt.	6.7	9.2
Heat of reaction, kJ/kg	345.0	629.9
∆Coke, wt.%	0.5	0.8

witnesses the range of variation of variables, which affect FCC unit performance, registered during 194 days of LNB FCC unit operation, when the above-mentioned catalysts and VGOs were employed.

In order to investigate which variables from Table 9 are statistically meaningful related to the FCC conversion, ICrA evaluation was used.

Table 10. μ—values from the ICrA evaluation of variables of FCC and H-Oil commercial units.

μ	MA	FCCU FR	%H-Oil VGO	H-Oil WABT	H-Oil FR	H-Oil FCC SLO, %	ROT	CFT	TRG Dense	TRG Dilute	H in Coke	Coke Yield	CTO	Heat of Reaction	∆ Coke	H-Oil VGO D15	FCC Feed D15	FCC Feed N	FCC conv
MA	1.00	0.49	0.43	0.35	0.49	0.53	0.50	0.46	0.58	0.58	0.39	0.55	0.51	0.58	0.56	0.49	0.47	0.43	0.41
FCCU FR	0.49	1.00	0.43	0.31	0.73	0.41	0.37	0.55	0.53	0.49	0.54	0.32	0.29	0.40	0.51	0.41	0.41	0.43	0.49
H-Oil VGO	0.43	0.43	1.00	0.43	0.52	0.56	0.39	0.35	0.58	0.58	0.49	0.53	0.34	0.38	0.61	0.58	0.78	1.00	0.25
H-Oil WABT	0.35	0.31	0.43	1.00	0.43	0.44	0.38	0.34	0.48	0.49	0.32	0.42	0.26	0.41	0.50	0.60	0.54	0.43	0.18
H-Oil FR	0.49	0.73	0.52	0.43	1.00	0.49	0.40	0.53	0.58	0.54	0.52	0.36	0.25	0.37	0.56	0.54	0.52	0.52	0.41
H-Oil FCC SLO,%	0.53	0.41	0.56	0.44	0.49	1.00	0.46	0.39	0.67	0.67	0.40	0.65	0.45	0.57	0.68	0.82	0.70	0.56	0.30
ROT	0.50	0.37	0.39	0.38	0.40	0.46	1.00	0.47	0.57	0.55	0.37	0.46	0.50	0.45	0.45	0.44	0.42	0.39	0.39
CFT	0.46	0.55	0.35	0.34	0.53	0.39	0.47	1.00	0.43	0.44	0.48	0.31	0.41	0.53	0.40	0.38	0.33	0.35	0.62
TRG dense	0.58	0.53	0.58	0.48	0.58	0.67	0.57	0.43	1.00	0.91	0.36	0.62	0.29	0.57	0.84	0.68	0.68	0.58	0.20
TRG dilute	0.58	0.49	0.58	0.49	0.54	0.67	0.55	0.44	0.91	1.00	0.35	0.64	0.31	0.60	0.84	0.68	0.68	0.58	0.20
H in coke	0.39	0.54	0.49	0.32	0.52	0.40	0.37	0.48	0.36	0.35	1.00	0.27	0.45	0.33	0.34	0.39	0.44	0.49	0.55
Coke yield	0.55	0.32	0.53	0.42	0.36	0.65	0.46	0.31	0.62	0.64	0.27	1.00	0.59	0.69	0.65	0.62	0.60	0.53	0.31
CTO	0.51	0.29	0.34	0.26	0.25	0.45	0.50	0.41	0.29	0.31	0.45	0.59	1.00	0.61	0.27	0.38	0.34	0.34	0.61
Heat of reaction	0.58	0.40	0.38	0.41	0.37	0.57	0.45	0.53	0.57	0.60	0.33	0.69	0.61	1.00	0.56	0.55	0.47	0.38	0.42
∆ Coke	0.56	0.51	0.61	0.50	0.56	0.68	0.45	0.40	0.84	0.84	0.34	0.65	0.27	0.56	1.00	0.71	0.71	0.61	0.18
H-Oil VGO D15	0.49	0.41	0.58	0.60	0.54	0.82	0.44	0.38	0.68	0.68	0.39	0.62	0.38	0.55	0.71	1.00	0.77	0.58	0.22
FCC Feed D15	0.47	0.41	0.78	0.54	0.52	0.70	0.42	0.33	0.68	0.68	0.44	0.60	0.34	0.47	0.71	0.77	1.00	0.78	0.17
FCC feed N	0.43	0.43	1.00	0.43	0.52	0.56	0.39	0.35	0.58	0.58	0.49	0.53	0.34	0.38	0.61	0.58	0.78	1.00	0.25
FCC conv	0.41	0.49	0.25	0.18	0.41	0.30	0.39	0.62	0.20	0.20	0.55	0.31	0.61	0.42	0.18	0.22	0.17	0.25	1.00

and

Table 11. υ—values from the ICrA evaluation of variables of FCC and H-Oil commercial units.

υ	MA	FCCU FR	%H-Oil VGO	H-Oil WABT	H-Oil FR	H-Oil FCC SLO, %	ROT	CFT	TRG Dense	TRG Dilute	H in Coke	Coke Yield	CTO	Heat of Reaction	∆ Coke	H-Oil VGO D15	FCC Feed D15	FCC Feed N	FCC Conv
MA	0.00	0.45	0.49	0.37	0.45	0.42	0.37	0.49	0.37	0.37	0.56	0.39	0.44	0.37	0.38	0.46	0.46	0.49	0.51
FCCU FR	0.45	0.00	0.48	0.40	0.20	0.53	0.50	0.40	0.42	0.46	0.40	0.62	0.66	0.54	0.44	0.53	0.52	0.48	0.42
%H-Oil VGO	0.49	0.48	0.00	0.27	0.40	0.37	0.46	0.58	0.35	0.35	0.44	0.38	0.60	0.55	0.31	0.34	0.14	0.00	0.65
H-Oil WABT	0.37	0.40	0.27	0.00	0.29	0.28	0.32	0.38	0.24	0.23	0.39	0.29	0.46	0.30	0.22	0.13	0.18	0.27	0.53
H-Oil FR	0.45	0.20	0.40	0.29	0.00	0.46	0.47	0.42	0.37	0.41	0.42	0.57	0.70	0.58	0.39	0.40	0.40	0.40	0.50
H-Oil FCC SLO,%	0.42	0.53	0.37	0.28	0.46	0.00	0.42	0.57	0.30	0.30	0.57	0.30	0.52	0.39	0.28	0.15	0.24	0.37	0.63
ROT	0.37	0.50	0.46	0.32	0.47	0.42	0.00	0.41	0.31	0.32	0.51	0.41	0.38	0.43	0.43	0.43	0.44	0.46	0.46
CFT	0.49	0.40	0.58	0.38	0.42	0.57	0.41	0.00	0.54	0.52	0.48	0.64	0.55	0.43	0.55	0.57	0.61	0.58	0.31
TRG dense	0.37	0.42	0.35	0.24	0.37	0.30	0.31	0.54	0.00	0.06	0.61	0.33	0.69	0.39	0.12	0.28	0.26	0.35	0.74
TRG dilute	0.37	0.46	0.35	0.23	0.41	0.30	0.32	0.52	0.06	0.00	0.61	0.31	0.65	0.36	0.12	0.27	0.26	0.35	0.73
H in coke	0.56	0.40	0.44	0.39	0.42	0.57	0.51	0.48	0.61	0.61	0.00	0.68	0.51	0.63	0.62	0.56	0.50	0.44	0.38
Coke yield	0.39	0.62	0.38	0.29	0.57	0.30	0.41	0.64	0.33	0.31	0.68	0.00	0.36	0.26	0.30	0.33	0.32	0.38	0.61
CTO	0.44	0.66	0.60	0.46	0.70	0.52	0.38	0.55	0.69	0.65	0.51	0.36	0.00	0.35	0.69	0.58	0.61	0.60	0.33
Heat of reaction	0.37	0.54	0.55	0.30	0.58	0.39	0.43	0.43	0.39	0.36	0.63	0.26	0.35	0.00	0.40	0.41	0.47	0.55	0.51
∆ Coke	0.38	0.44	0.31	0.22	0.39	0.28	0.43	0.55	0.12	0.12	0.62	0.30	0.69	0.40	0.00	0.24	0.22	0.31	0.75
H-Oil VGO D15	0.46	0.53	0.34	0.13	0.40	0.15	0.43	0.57	0.28	0.27	0.56	0.33	0.58	0.41	0.24	0.00	0.17	0.34	0.70
FCC Feed D15	0.46	0.52	0.14	0.18	0.40	0.24	0.44	0.61	0.26	0.26	0.50	0.32	0.61	0.47	0.22	0.17	0.00	0.14	0.74
FCC feed N	0.49	0.48	0.00	0.27	0.40	0.37	0.46	0.58	0.35	0.35	0.44	0.38	0.60	0.55	0.31	0.34	0.14	0.00	0.65
FCC conv	0.51	0.42	0.65	0.53	0.50	0.63	0.46	0.31	0.74	0.73	0.38	0.61	0.33	0.51	0.75	0.70	0.74	0.65	0.00

present μ and υ values obtained from the ICrA assessment. From all investigated variables, five were identified by ICrA to have a statistically meaningful relation to the FCC conversion. These variables are as follows: ΔCoke (μ = 0.18; υ = 0.75), FCC feed density (μ = 0.17; υ = 0.74), H-Oil VGO density (μ = 0.22; υ = 0.70), regenerator dense phase temperature (μ = 0.20; υ = 0.74), and regenerator dilute phase temperature (μ = 0.20; υ = 0.73). All of them are in negative consonance, implying that their magnification will be associated with a reduction in the FCC conversion. The variables FCC feed density and H-Oil VGO density concern the effect of feed quality on FCC conversion, as exemplified in Equation (3). Whereas the variable ΔCoke is a function of both feed quality and catalyst coke selectivity [52,53,54]. The regression analysis of the data for 194 days of LNB FCC unit operation confirmed that ΔCoke depends on two variables related to FCC feed quality—density, and to catalyst characteristics—micro-activity, of the equilibrium catalyst. Equation (4) displays the developed ΔCoke correlation.

$$\begin{gathered} ∆Coke=-3.2455+3.4659\times FCCfeed{d}_{15}+0.009245\times \\ ECATMicroactivity R = 0.765, \mathrm{standard} \mathrm{deviation} = 0.027 \mathrm{wt}.\% \end{gathered}$$

(4)

where

ΔCoke = the difference between coke on spent catalyst and coke on the regenerated catalyst in the commercial FCC unit, wt.%;

FCC feed d₁₅ = density at 15 °C of the combined FCC unit feed, g/cm³;

ECAT Micro-activity = Micro-activity of equilibrium catalyst from the FCC unit, wt.%. (This is the conversion of a standard feed, measured in a laboratory ACE FCC unit on an equilibrium catalyst sample at a catalyst-to-oil ratio of 4.0 wt./wt., and reaction temperature of 527 °C)

The regenerator temperatures are dependent on the ΔCoke (μ = 0.84; υ = 0.12) and therefore their influence on the FCC conversion is accounted by the ΔCoke.

Regression of FCC feed crackability data, estimated by density and Equation (3), along with ΔCoke, against the FCC conversion led to the development of Equation (5).

$$\begin{gathered} FCCconversion=14.065+1.0266\times Crackability-31.091\times ∆Coke, \\ R = 0.864, \mathrm{standard} \mathrm{deviation} = 1.52 \mathrm{wt}.\% \end{gathered}$$

(5)

where

FCC conversion = conversion of material boiling above 210 °C in FCC unit, wt.%;

Crackability = estimated by Equation (3) conversion of FCC feed at the maximum gasoline yield observed in laboratory ACE FCC unit, wt.%.

For this dataset, the riser outlet temperature had no statistically meaningful effect on the FCC conversion. Laboratory cracking experiments in an ACE FCC unit with a feed consisting of 75% straight run VGO and 25% H-Oil VGO, having 73.0% crackability, and using catalyst F in the reactor, a temperature range of 526–550 °C, were used to elucidate the reason for not detecting any effect of reactor temperature on conversion in the commercial FCC unit. Figure 15 witnesses the variation of FCC feed conversion with catalyst to oil ratio changing for two reaction temperatures (526 and 550 °C). One can calculate from the data in Figure 15 that at each 10 °C reaction temperature alteration, the FCC conversion fluctuates with about 1.3 wt.%.

Having in mind that the fluctuation of reaction temperature for most of the FCC data presented in this work ranges between 540 and 550 °C, the effect of reactor temperature seems to be suppressed by the wider ambit of VGO feed crackability variation. The FCC feed crackability was mainly controlled by the percent of H-Oil VGO in the FCC feed and by the content of FCC slurry oil in the H-Oil feed, since the blended straight-run VGO crackability for the studied data varied in the narrow range of between 72.0 and 75.6 wt.%. A regression of these two variables against the FCC conversion led to the development of Equation (6).

$$\begin{gathered} FCCconversion=78.7-0.203\times \%HOilVGOinFCCfeed-0.343\times \\ \%FCCSLOinHOilfeed, R = 0.802, \mathrm{standard} \mathrm{deviation} = 1.80 \mathrm{wt}.\% \end{gathered}$$

(6)

where

%HOilVGO in FCC feed = the content of H-Oil VGO in FCC feed, wt.%;

%FCCSLO in H-Oil feed = the content of FCC slurry oil in H-Oil feed, wt.%.

It deserves mentioning here that the increase of equilibrium catalyst activity from 69.0 to 77.1 wt.% (see Table 9) does not give a positive effect on the commercial FCC unit conversion. Instead, it increases ΔCoke (see Equation (4)) that eventually leads to a reduction of the FCC conversion by 2.4 wt.% (see Equation (5)).

The 194 operating days of the FCC unit were employed to construct an ANN model predicting FCC conversion. Figure S7 exemplifies how the neural network training process has been accomplished to build the ANN commercial FCC vacuum gas oil feed conversion prediction model. The data in Figure S5 shows that the maximum number of epochs allocated for training was 5000, the time for the ANN training took 68 s, the desired mean squared error was set at $1\times {10}^{-6}$, as well as the achieved mean squared error was at $4.31\times {10}^{-6}$, and the training method used was that of Levenberg–Marquardt.

Figure 16 indicates graphs of the neural network training performance for prediction of commercial H-Oil hydrocracker vacuum residue net conversion. One can see from the data in Figure 17 the three types of data: training, testing, and validation. The graph shows that the achieved mean squared error was 0.0015754, which occurred at epoch 10. After epoch 10, the mean squared error did not keep on falling for validation, meaning that no further improvement in the model accuracy could be expected.

Figure 17 is a parity graph of predicted versus measured (target) FCC VGO feed conversion for the three types of data: training, testing, and validation. The correlation coefficients of the three types of data were found to be R = 0.99182 for training, R = 0.95393 for testing, R = 0.74011 for validation, and overall R = 0.89046. The overall data comprised all subsets of data for training (70% of all data), testing (20% off all data), and validation (10% of all data). The lower accuracy of the FCC conversion fit for the validation data is a result of quite inaccurate prediction of conversion of 100% H-Oil VGO (65% predicted versus 48.2 wt.% observed). There was only one point for 100% H-Oil VGO being a feed of the studied commercial FCC unit that was not included in the training of the ANN model. Therefore, a good FCC ANN model would require all possible variations to be present in the training and validation processes to a fairly large extent. The test set that included data not utilized in the training and validation demonstrated quite a good predictability with R = 0.9593, which is much better than the multiple correlation coefficient of regression Equation (5) (R = 0.864).

3. Materials and Methods

Seventeen distinct crude oils and two imported atmospheric residues with characteristics shown in Table S1 were processed in the LUKOIL Neftohim Burgas (LNB) refinery under study. Properties of the derived vacuum residues used as feedstocks for the ebullated bed vacuum residue are summarized in Table 1. Characteristics of the vacuum gas oils obtained from the 19 refinery feedstocks employed as feed for the LNB FCC unit are presented in Table 7. Features of FCC catalysts availed in the LNB FCC unit during the study are displayed in Table 8. Three low-sediment Ni–Mo-supported solid catalysts and liquid nano-dispersed HCAT catalysts [55] were used in the commercial ebullated bed vacuum residue hydrocracker during the research.

The process scheme of the LNB refinery where the fluid catalytic cracking and the ebullated bed vacuum residue H-Oil hydrocracking units were investigated is presented in Figure S1. The flow diagrams of both H-Oil and FCC units under study are shown in Figures S2 and S3. Two different fresh catalyst addition systems (cascade and parallel) in the commercial ebullated bed vacuum residue H-Oil hydrocracker were investigated in this research. Their simplified scheme is shown in Figure S6.

Six individual vacuum residues, different from those whose properties are shown in Table 1, were hydrocracked in a pilot plant with the aim to establish a quantified relation of vacuum residue properties to conversion level. The properties of these six vacuum residues and the process flow diagram of the pilot plant are given in Table S1 and Figure S4, respectively.

The FCC conversion was calculated by using Equation (7).

$$FCCconversion\left(wt.\%\right)={\displaystyle \frac{{FCCfeed}_{>210°C}-{FCCproduct}_{>210°C}}{{FCCfeed}_{>210°C}}}\times 100$$

(7)

where

FCC feed_>210°C = the amount of material boiling above 210 °C in FCC feed, t/h;

FCC product_>210°C = the amount of material boiling above 210 °C in FCC product, t/h.

Simulated distillation ASTM D 2887 [56] was used to measure the content of material boiling in both FCC feed and product.

The H-Oil net conversion was calculated by employing Equation (8).

$$HOilconversion(wt.\%)={\displaystyle \frac{{HOilFeed}_{540°C+}-{HOIlProduct}_{540°C+}}{{HOilFeed}_{540°C+}}}\times 100$$

(8)

where

HOilFeed_540°C+ = mass flow rate of the H-Oil feed fraction, boiling above 540 °C, determined by the high temperature simulated distillation method ASTM D 7169 [57] of the feed and multiplied by the mass flow rate of the feed in t/h;

HOilProduct_540°C+ = mass flow rate of the H-Oil product fraction, boiling above 540 °C, determined by high temperature simulated distillation method ASTM D 7169 [57] of the liquid product multiplied by the flow rate of the liquid product in t/h.

The balance conversion of commercial H-Oil unit is calculated by Equation (9).

$$HOilbalanceconversion(wt.\%)=100-VTB-H2S$$

(9)

where

VTB = yield of the vacuum tower bottom product in H-Oil hydrocracker, wt.%;

H2S = yield of hydrogen sulfide in H-Oil hydrocracker, wt.%.

Table 12. Methods used to measure oil properties.

Properties Measured	Standard Method
Density of oils, g/cm3	ASTM D5002 [58]
Sulfur of studied oils	ASTM D 4294 [59]
Asphaltene (C7, and C5) content, wt.%	ASTM D 6560 [60]
Conradson carbon content, wt.%	EN ISO 10,370 [61]
Nitrogen content, wt.%	ASTM D 5291 [62]
Nickel, ppm	IP 501 [63]
Vanadium, ppm	IP 501 [63]
Sodium, ppm	IP 501 [63]
Iron, ppm	IP 501 [63]
SARA composition	In house method [64]

summarizes the methods employed to characterize the investigated oils.

Three knowledge discovery in databases (KDD) techniques: intercriteria (ICrA) and regression analyses, and artificial neural network modeling were utilized in this research. ICrA evaluation of the data generated by the operation of the FCC and the ebullated bed vacuum residue hydrocracking units was implemented to search for the relations between feed quality, catalyst characteristics, and the unit operating conditions to the conversion level. The regression analysis was performed with two purposes: (1) to model conversion level as a function of feed quality, catalyst characteristics, and unit operating conditions; and (2) to define the combination of variables mentioned before, which have statistically meaningful effects on conversion level in both FCC and H-Oil commercial units.

The essence of ICrA has been described in detail in [65]. ICrA has been developed on the basis of intuitionistic fuzzy sets and index matrices, and two variables μ and υ are calculated by using specialized software freely available as open source from https://intercriteria.net/software/ (accessed on 23 August 2024) and explained in [66,67]. Instead, the conception “correlation” between investigated variables ICrA applies the notions “positive consonance”, “negative consonance”, and “dissonance”. For industrial objects, which are featured by a relatively strong “noise” brought about by different disturbances of the process, the meaning of μ = 0.70–1.00 and υ = 0–0.30 represents a region of statistically meaningful positive consonance, while at μ = 0–0.30 and υ = 0.70–1.00, an area of statistically meaningful negative consonance is obtained. All other cases are regarded as dissonance. We decided to apply ICrA because it can register the presence of statistically meaningful relations that are both linear and non-linear, while conventional correlation analysis uncovers the presence only of linear relations.

The artificial neural network (ANN) made in this research was distinguished by the fact that the outputs of each layer were connected to the inputs of the next layer. Thus, the ANN had no feedback, making it a feedforward neural network. Its structure consisted of a first hidden layer with 123 neurons; a second layer with 42 neurons; a third layer with 16 neurons; a fourth layer with 10 neurons; a fifth layer with 8 neurons; and a sixth layer with only 1 neuron. All the data were separated into three parts: training, validation, and testing. The training set is utilized to train the model, the validation set helps in model selection and hyperparameter tuning, and the test set assesses the ANN execution on data not embraced in the training or validation data. When the mean squared error calculated on the validation data starts to increase, it is deemed that the ANN training process has to be stopped. The selection of the structure of ANN was advised by the complexity of the process. By magnification of the number of neurons in the critical first layer, it was expected to achieve the best result. However, further increasing the number of neurons led to an accumulation of the error that accompanied each neuron. In this way, it was experimentally found the balance in the numbers of neurons in each layer that resulted in the lowest mean squared error. At the input of the neural network for the ebullated bed vacuum residue H-Oil hydrocracker, the following data were provided: the contents in the feed of Conradson carbon (wt.%), sulfur (wt.%), nitrogen (wt.%), saturates (wt.%), aromatics (wt.%), resins (wt.%), C₇-asphaltenes (wt.%), percent of vacuum residue in H-Oil feed (boiling above 540 °C), feed reactivity (%), weight average bed temperature (WABT, °C), and the liquid hourly space velocity (LHSV, h⁻¹). The input of ANN for the FCC unit was fed with the following data: equilibrium catalyst micro-activity (%), FCC unit feed rate (t/h), percent of H-Oil vacuum gas oil (VGO) in the FCC feed, WABT of H-Oil reactors (°C), H-Oil feed rate (t/h), content fluid catalytic cracking slurry oil (FCC SLO) in H-Oil feed (wt.%), riser outlet temperature (°C), combined feed temperature (°C), ∆Coke (wt.%), H-Oil VGO density at 15 °C (g/cm³), FCC combined feed density at 15 °C (g/cm³), and FCC feed nitrogen content (wt.%).

The training, validation, and test sets of the data used by the ANN were divided as follows: 70% for training; 20% for testing; and 10% for validation. A random splitting algorithm was employed for the data allocation.

4. Conclusions

The correlation developed to relate vacuum residue properties to vacuum residue conversion in a laboratory ebullated bed hydrocracking plant was found to correlate with the performance of the commercial EBVRHC unit. It provides a quantitative estimation of feed reactivity in the ebullated bed vacuum residue hydrocracking, and together with the commercial unit operating conditions, can be used to predict the commercial hydrocracker vacuum residue conversion. This correlation revealed that the vacuum residual oils that were hydrocracked in the commercial EBVRHC unit under study significantly differ in their reactivity. The single application of ICrA to evaluate commercial EBVRHC unit data with the aim to detect statistically meaningful relations is not sufficient, and its combination with regression analysis enables discovering the combination of variables that are related to the target variable, which individually have been found to have statistically meaningless relation to the target variable. The application of ANN modeling presents an opportunity to improve the accuracy of the commercial EBVRHC unit performance prediction over that achieved by using regression analysis.

It was found that metal sulfides coming from the vacuum residue form a non-homogeneous layer on the outer surface of the catalyst in the second reactor of the commercial EBVRHC unit. This layer seems to be broken in the first reactor when the cascade system for catalyst addition in the first reactor is employed, resulting in black powder generated there. This mechanism of black powder formation in the first reactor is backed up by the same ratio of metals found in the black powder and that in the feed of the commercial EBVRHC unit. Thus, the replacement of the cascade with a parallel system to add catalyst in both reactors seems to provide an option to exclude the second reactor spent catalyst from entering the first reactor and avoid the formation of black powder in the first reactor. In this way, 130 days of far from optimal performance of the EBVRHC unit can be dodged.

Similar to the commercial EBVRHC unit, the commercial FCC unit performance also demonstrated a strong dependence on the vacuum gas oil crackability that can be predicted on the basis of a correlation developed by using laboratory cracking tests. The catalyst activity, which, along with FCC feed density, correlated with ΔCoke, the other significant factor influencing FCC conversion, exhibited a negative impact on conversion because it enhanced ΔCoke. The ANN model of prediction of vacuum gas oil conversion in the commercial FCC unit, like the case with the EBVRHC unit, demonstrated a higher accuracy of conversion prediction than the regression model. It should be noted here, however, that the ANN model significantly overpredicted the conversion of 100% H-Oil VGO; that was the only data point of its kind in a database of 194 points, indicating that a good FCC ANN model would demand all possible variations of the input data to be present in the training and validation processes to a fairly large extent.

The ANN modeling technique demonstrates better conversion predictability in both commercial EBVRHC and FCC units compared to the regression technique, which suggests that in the future the ANN modeling can be used as a base for building an advanced process control of these process units.