COKE MORPHOLOGY BY IMAGE SEGMENTATION

Abstract

The present invention is directed to a method for the prediction of coke morphology from feed characteristics using cross-polarized light optical microscopy, image segmentation, and statistical analysis.

Claims

1. A method for the prediction of coke morphology from feed characteristics comprising a) Performing a microcarbon test (MCRT) on a coker feed, ASTM D 4530 to produce a MCRT coke sample, b) Using cross-polarized light optical microscopy at 500× to produce a photo of the MCRT coke sample, c) Using machine learning segmentation software to produce a segmented output file that comprises a partitioned image with multiple segments, d) Determining structural parameters of output file by applying statistical analysis weighted by area, e) Correlating the resulting statistical analysis to a coke morphology.

2. The method of claim 1 wherein the image segmentation wherein the coke morphology consists of shot, sponge and transitional coke.

3. The method of claim 1 wherein the structural parameter is Feret maximum calculated by statistical analysis.

4. The method of claim 1 wherein the structural parameter is number of particles less than or equal to 10 μm calculated by statistical analysis.

5. The method of claim 4, wherein there is a correlation of area, ferret max and particles less than or equal to 10 μm to % HHI/MCRT ratio, Asphaltene solubility parameter, Asphaltene peptizability.

6. The method of claim 1 wherein the structural parameter is area calculated by statistical analysis.

7. The method of claim 1 wherein the feed is selected from the group consisting of blends of petroleum derived feedstocks, virgin and/or previously converted feeds, low percentage of distillable materials, high sulfur and nitrogen feeds, high metal containing feeds.

8. The method of claim 5 wherein the metal in the high metal feed is selected from the group consisting of vanadium and nickel.

9. The method of claim 1 wherein the image segmentation steps yields at least 1000 individual counts per image.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0012] FIG. 1 is a diagram of the most common methodology to determine coke morphology based on the combination of Microcarbon Residue Test (MCRT) on a coker feed followed by cross-polarized light optical microscopy (CPL-OM) of the MCRT cokes.

[0013] FIG. 2 is a micrograph of an MCRT coke sample using cross-polarized light optical microscopy.

[0014] FIG. 3 is a plot of the mean-particle size determined using the method described in FIG. 1 vs. the percentage of hot heptane insolubles (% HHI)/% MCRT ratio for a series of nine Delayed Coking feeds.

[0015] FIG. 4 is a micrograph of a known sponge coke sample using cross-polarized light optical microscopy.

[0016] FIG. 5 is a diagram of the claimed method described herein to determine coke morphology based on the combination of Microcarbon Residue Test (MCRT) on a coker feed, cross-polarized light optical microscopy of the MCRT cokes, image segmentation, and statistical analysis.

[0017] FIG. 6 is a photo of a typically segmented micrograph obtained from processing the one depicted in FIG. 2 using a segmentation software such as the Zeiss Zen™ Intellesis™ module.

[0018] FIG. 7 is a plot of the average Feret maximum (μm) using the claimed methodology described in FIG. 5 vs. the % HHI/% MCRT ratio for a series of four-teen Delayed Coker feeds.

[0019] FIGS. 8 & 9 are plots of the average area and percentage of particles with Feret maximum lower than 10 μm vs. the % HHI/% MCRT ratio for a series of 14 Delayed Coking feeds. As seen, a smaller average area and a higher percentage of particles with Feret maximum lower than 10 μm can be found in shot cokes than the sponge coke counterparts. These correlations have relatively weak R.sup.2 (0.45-0.55), but the tendency of the data is clear.

[0020] FIGS. 10 and 11 are plots of the average Feret maximum and average area of particles determined using the segmentation method vs. the solubility parameter for a series Delayed Coking feeds. As seen, correlation factors in the 0.67-0.75 were found. Thus, feeds with low average Feret maximum and average areas have a higher tendency to form shot coke.

[0021] FIG. 12 is a plot of the average area of particles determined using the method of the invention described herein (FIG. 5) vs. asphaltene peptizability (Pa) for a series of 12 Delayed Coking feeds.

DETAILED DESCRIPTION OF THE INVENTION

[0022] The most common methodology (FIG. 1) to determine coke morphology is the combination of Microcarbon Residue Test (MCRT) on a coker feed followed by cross-polarized light optical microscopy (CPL-OM) of the MCRT cokes. CPL-OM uses a lambda plate and is recommended for specimens with very low reflectivity (such as coke and coal samples). Because polarized light converts contrast from gray to different colors, this technique exposes significant features of the coke that otherwise would not be captured. The methodology is shown in FIG. 1 involves taking several photographs of the MCRT cokes at 500× (see typical CPL-OM micrograph in FIG. 2), determining the mean-particle size manually or using an image processing software such as QImaging (http://www.ddisoftware.com/qimage-u/index.html) of −300-400 particles and correlating with the coke morphology. By this way, shot-coke-forming feeds produce cokes with a less anisotropic mosaic structure of 1-10 μm whereas sponge coke-generating feeds lead to cokes with highly anisotropic (ordered) 10-60 μm flow domains. As described herein, image segmentation can effectively predict coke morphology in VR Blends.

[0023] Herein is described a methodology based on CPL-OM, image segmentation, and statistical analysis to predict coke morphology comprising shot, sponge, and transitional coke in Delayed Coking from feed characteristics. Image segmentation is the process of partitioning a digital image into multiple segments to simplify and/or change the representation into something easier to analyze statistically. As shown in FIG. 5., the new methodology involves taking several photographs of the MCRT cokes at 500× by CPL-OM, obtaining the segmented output file using a segmentation software such as Zeiss Zen™ Intellesis™ module, Manser, R., Elsässer, R., Doting, V., ZEISS ZEN Intellesis Machine Learning Approaches for Easy and Precise Image Segmentation, July 2018, Carl Zeiss Microscopy GmbH, Germany) determining several structural parameters by applying statistical analysis weighted by area using Microsoft Excel, and correlating those parameters with the coke morphology.

[0024] Thus, an embodiment of the invention as supported herein comprises:

[0025] 1) Performing a microcarbon test (MCRT) on a coker feed, ASTM D 4530 to produce a MCRT coke sample,

[0026] 2) Using cross-polarized light optical microscopy at 100×, 200× or 500×, preferably 500×, to produce a photo of the MCRT coke sample,

[0027] 3) Using machine learning segmentation software to produce a segmented output file that comprises a partitioned image with multiple segments,

[0028] 4) Determining structural parameters of output file by applying statistical analysis weighted by area,

[0029] 5) Correlating the resulting statistical analysis to a coke morphology.

[0030] A further embodiment of the invention is the determination of critical structural parameters comprises the calculation of the weighted distributions of specific characteristics of the particles identified by image segmentation. The probability density function ƒ(x.sub.i) of a weighted random variable or particle characteristic x.sub.i is given by:

[00001]$f\left(x_i\right)=\frac{x_i.A_i}{\mu }$

Where x.sub.i is the i-particle characteristic such as maximum ferret, area, and elongation, A.sub.i represents the area of the i-particle, n is the number of particles and μ is given by:

[00002]$\mu =\underset{i=1}{\overset{n}{.Math.}}{x}_i{A}_i$

The weighted average x is calculated as:

[00003]$\overline{x}=\frac{\underset{i=1}{\overset{n}{.Math.}}{x}_i{A}_i}{\underset{i=1}{\overset{n}{.Math.}}{A}_i}$

The cumulative distribution function is given by:

[00004]$F\left(x_m\right)=\frac{\underset{i=1}{\overset{m}{.Math.}}{x}_i{A}_i}{\mu }$

FIG. 6 shows a typically segmented micrograph obtained from processing the one depicted in FIG. 2. Significant features can be observed by segmentation that cannot be seen in a cross polarized light optical microscope.

[0031] An embodiment of the invention is the use of machine learning for image segmentation. The main advantages versus conventional methodologies (FIG. 1) are: a much larger set of particles are counted, equal to or greater than 1000 individual particles, so an improved repeatability is obtained throughout the whole range of particle sizes, less data dispersion and better correlations are obtained, and more structural parameters are correlated with feed characteristics. The use of a large data set yields meaningful, repeatable, and highly reliable statistics enabling one of skill in the art to more accurately correlate coke morphology to other feed properties. Furthermore, a single segmented model can be applied to different micrographs of coke from the same feed taken under the same conditions.

[0032] A further embodiment of the invention is utilizing any commercial software to perform image stitching. Image stitching is used to combine multiple photographic images with overlapping fields of view to generate a single, segmented high-resolution image. In this way, a much larger set of particles are counted. Image stitching is especially important when the individual grains of coke are greater than 20 μm like those found in sponge coke.

[0033] A further embodiment of the invention is the use of a high-resolution camera no less than 5 megapixels, preferably 12 megapixels, to capture the cross-polarized light optical microscopy images. In this way, higher resolution can be achieved especially for individual grains of coke that are smaller than 10 μm like those found in shot coke.

[0034] A further embodiment of the invention is predicting coke morphologies from blends of petroleum derived feedstocks, virgin and/or previously converted feeds, low percentage of distillable materials, high sulfur and nitrogen feeds, high metal containing feeds comprising of vanadium and nickel.

[0035] A further advantage to the method described herein is that the particles located on the edges of the segmented micrographs can be easily removed to improve repeatability. More than 2000 reflectors are analyzed from 2 representative images per feed. Following the image segmentation, the Zeiss Zen™ Intellesis™ software can perform a variety of image analysis techniques to yield statistics of more than 90 morphological parameters from each image. For this invention, the parameters selected were, but not limited to, the average Feret maximum, Feret minimum, particle's individual identification, region class color name, compactness, circularity, roundness, average area, percentage of particles with Feret lower than 10 μm, and elongation. It is important to mention that Feret is the distance of two tangent lines to a contour of the particle and is considered a measurement of the particle size.

[0036] Other structural parameters obtained by the segmentation method (FIG. 5) could be used to predict the coke morphology. This data is not available using the conventional methodology (FIG. 1). For example, FIGS. 8 and 9 show the average area and percentage of particles with Feret maximum lower than 10 μm vs. the percentage of hot heptane insolubles (% HHI)/% MCRT ratio for a series of 14 Delayed Coking feeds. As seen, smaller average area and a higher percentage of particles with Feret maximum lower than 10 μm can be found in shot cokes than the sponge coke counterparts. These correlations have relatively weak R.sup.2 (0.45-0.55), but the tendency of the data is clear.

[0037] Similarly, other feed characteristics can be correlated to the structural parameters determined by the method described in this invention (FIG. 5) and used to predict coke morphology for Delayed Coking. Siskin et al. reported that asphaltenes with higher solubility parameters favor phase separation from the hydrocarbon matrix and lead to shot coke formation. FIGS. 10 and 11 show the average Feret maximum and average area of particles determined using the segmentation method vs. the solubility parameter for a series Delayed Coking feeds. As seen, correlation factors in the 0.67-0.75 were found. Thus, feeds with low average Feret maximum and average areas have a higher tendency to form shot coke.

[0038] It is known in the state of the art that unstable feeds favor the formation of shot cokes, U.S. Pat. No. 7,803,627, US 2012/0298553. These reports give support to the hypothesis that asphaltene stability and coke morphology are linked. By stability measurements using transmittance to detect the flocculation onset, the peptizability (Pa) can be determined. As seen in FIG. 12, the average area of particles determined using this invention (FIG. 5) is directly proportional to Pa for a series of twelve Delayed Coking feeds. Thus, lower peptizability, more unstable the feed becomes and therefore, a higher tendency to form shot coke. This correlation has a relatively small R.sup.2 (0.42), but the trend of the data is clear.