Alain Espinosa, Michel Sanchez, Sebastien Osta, Christian Boniface, Jose Gil
BP France
Lavera, France
Andre Martens, Bernard Descales, Didier Lambert
BP Chemicals
Lavera
Marc Valleur
Technip
Paris
A near-infrared (NIR), advanced control system has been controlling the gasoline blender at BP France's Lavera refinery for more than 2 years. The blender produces 1 million metric tons/year gasoline (about 21,000 b/d) using multivariable blend-control software in a closed-loop mode.
BP's system uses a centralized NIR spectrometer fitted with optical fibers that carry NIR radiation between the instrument and process streams. The analyzer monitors the product stream from the blender and transmits data to the control software, which adjusts the component flow rates to meet target specifications.
The system computes blending indices using an innovative approach based on the NIR response of the individual basestocks.
NIR's high level of repeatability coupled with robust in-house models developed through more than 15 years of experience, has improved the reliability of blending operations. Octane giveaway has been reduced by 0.3 points, saving BP an estimated $2 million/year.
NIR ANALYSIS
NIR has a number of advantages over conventional octane-measurement methods and is finding widespread use in the process industries. Among NIR's advantages is speed--less than 1 min is required to obtain a spectrum. The technique has a repeatability of 0.05%, and can be applied to liquids, waxes, and solids without sample dilution.
Whether used as a dedicated instrument, or in combination with optical fibers and a multiplexer, NIR is an ideal technique for on-line analysis. The NIR spectrum ranges from 800 to 2,500 nm. The sample spectra contain information about the whole molecule, and are condensed through complex overlapping of combination and overtone bands.
Because of their complexity, NIR spectra almost always require computer analysis. The most common analysis method is correlation of NIR absorbencies with physical and chemical properties through appropriate statistical treatments. This approach makes possible the measurement of several properties simultaneously from a single spectrum.
BLENDING OPTIMIZATION
BP has been developing applications for NIR technology since 1974, and holds 23 patents. NIR on-line systems have been installed at the Lavera refinery in such applications as: steam cracker optimization, crude oil distillation process control, and blending optimization for leaded and unleaded motor spirits.
Typical steps in a gasoline-blending operation are: basestocks analyses for determining blending indices, blending-order calculation, and blender monitoring.
The weaknesses of the blending process are the inaccuracy of blending indices and the difficulty in controlling set points for component flow rates. Because of these sources of error, re- blending frequently is required to either bring the product within specifications or reduce giveaway.
The determination of blend indices is linked directly to a refiner's blending policy. A discrepancy can occur if the usual proportion of basestocks is changed, or if a new basestock is introduced. And the more complex the pool, the higher the probability of error.
Changes in basestocks are common in blending operations. The shutdown of a unit or the import of new basestocks are not exceptional events, and they necessitate correction of the blending indices.
For example, the use of oxygenates is a great source of error because of the strong variation of their response, depending on the nature of the gasoline pool. Moreover, blending indices for octane number are determined from the cooperative fuel research (CFR) engine test, which lacks accuracy and has a long response time.
These factors make gasoline blending a difficult operation to optimize, while reblending is time-consuming and, therefore, costly.
LAVERA PROJECT
BP's Lavera complex comprises chemical plants and a sophisticated refinery with major units including: crude distillation, vacuum distillation, reforming, isomerization, fluid catalytic cracking, hydrocracking, and visbreaking (Fig. 1). In addition, a selective hydrogenation unit eliminates dienes in the steam-cracked spirit received from an associated BP Chemicals plant.
The gasoline pool comprises more than 16 basestocks (Table 1). Three main gasoline grades are produced: M98 unleaded (98 RON, 88 MON), M95 unleaded (95 RON, 85 MON), and M97 leaded (97 RON with 0.13 g/l. lead).
BP's main goals for the NIR-based blending system were:
- Minimization of quality giveaway
- Optimization of blend recipes
- Increase in blender flexibility by avoiding reblends * Reduction of future storage-capacity requirements for blended products.
The complex nature of refinery hydrocarbon mixtures requires the use of advanced chemometric techniques to develop statistical models that relate properties such as octane number and vapor pressure to NIR spectral patterns. In this case, the keys to robust measurement are BP's 15 years of NIR expertise and its data base, built from large sets of crude and refinery blending streams. The data base includes more than 12,000 NIR spectra from nine instruments and seven refineries.
The on-line NIR system comprises:
- A spectrometer measuring in the 2,000-2,500 nm wavelength range
- Zirconium fluoride optical fibers
- A measurement cell with a path length of 500 T
- A computer for data analysis and instrument control.
Figs. 2 and 3 show, respectively, a chart of the gasoline-blending operations and a flow scheme of the blend-control system.
The first step in the blending process is the measurement of the basestock NIR spectra. These provide blending indices for vapor pressure, distillation points, volatility (a function of RVP and % distilled at 700 C.), and research and motor octane number, clear and leaded, for the three grades of finished products.
This information is transferred through computer link to the planning department, where blending-order calculations are performed based on these NIR blending indices. Data are then transferred to the off sites department for incorporation in the blender control system.
During a blending-operation, the blender is monitored by Technip's multivariable control software, Anamel. This software provides set points for the component control valves, according to the blending order, and integrates the blend qualities from on-line NIR analyses. In these analyses, RON and MON are measured every 45 sec, as are Rvp, density, volatility, and distillation points.
This system makes it possible to accumulate N measurements during a given time period, and reduce by a factor of checkN the discrepancy between the measurements and the true value. The response time of NIR spectroscopy thus gives a unique and decisive advantage over knock-engine measurement.
The blending software can manipulate as many as eight components for a given blend and control a maximum of five product qualities simultaneously. Some components can have fixed set points. The system includes a function that minimizes the cost of the formulation while fulfilling the constraints.
The finished product from the tank is analyzed by NIR in the laboratory. In addition to the qualities measured by the on-line analyzers, laboratory analysis provides the concentrations of benzene, methyl tertiary butyl ether, oxygen, sulfur, and gums. In all, 14 properties are obtained directly from a single NIR measurement.
- Addition of NIR on-line instrumentation (Fig. 4)
- Automatic transmission of laboratory NIR spectra into a laboratory data base
- Modification of blending calculation program with gasoline blending indices determined by NIR (implemented on a DCS station)
- Implementation of Anamel software on gasoline blender
- Integration of Anamel software with the existing off site information and control system.
BP METHODOLOGY
BP aims to create robust NIR models and to be safe in the treatment of outliers. The company develops its own software specific to local blending recipes and based on its own industrial needs, thus producing more-reliable models than those from commercial statistical packages.
The method employed to produce these models is based on early detection of outliers by pattern-recognition techniques, whereas other property predictions are achieved through a specific modeling approach.
DATA TREATMENT
Before moving into the on-line analysis phase, the NIR results must be validated in the quality-control laboratory. This validation step is crucial for any new technique and requires comparing the results obtained from the standard method with those obtained from the new method.
Over a 6-month period, NIR predictions and ASTM-method results were compared for: unleaded and leaded MON and RON, Rvp, distillation, and volatility. The results show that the accuracy of the properties predicted from the NIR spectra is at least as good as that from standard ASTM methods (Fig. 5).
Comparing the deviation between two CFR engines, the total standard deviation (TSD) for ASTM octane number measurements was obtained using the standard deviation given by the ASTM curve.1 Thus, the standard deviation values for RON and MON have been retained (respectively, 0.25 and 0.4, which correspond to 97-98 octane 1).
The TSD for the NIR method is obtained by comparing it with the standard method. For octane number, for example, the TSD has been calculated from more than 100 experimental measurements. NIR appears to produce better octane numbers--both RON and MON, leaded and unleaded--than does the ASTM method. NIR's TSDs for other properties are roughly the same as from the ASTM methods.2
These results are attributable to the good repeatability of the NIR technique and to the stability and robustness of BP's models. In addition to NIR's measurement accuracy, the analyzer itself has proved robust and stable, with a consequent reduction in both operating and maintenance costs.
SYSTEM ASSESSMENT
To validate the system installed on the blender, the NIR laboratory and on-line measurements were compared. Figs. 6a and 6b show the distribution of the differences between the two sets of measurements over 6 months of operation.
These results show strong consistency, with a standard deviation of 0.12 and 0.15 for, respectively, MON and RON, and a maximum deviation of, respectively, 0.3 and 0.31. The maximum deviations can be explained by the possible heterogeneity of the finished-product tank, and by the laboratory NIR's repeatability of 0.1 octane number.
Comparing the same values obtained from the integrated on-line analysis with the target fixed by the blending order, the maximum deviation is found to be 0.15 octane number, while the standard deviation is 0.05 (Fig. 7).
This small deviation is caused mainly by the fact that several basestock qualities are controlled with minimum degrees of freedom--typically one. In fact, the Anamel system works by reaching the quality targets and minimizing the differences between the initial blending order and the actual formula. This leads, in this case, to a maximum deviation of 0.15.
It should be noted that, before the on-line system and new NIR blending indices were implemented, the difference between the blending order and the final measured value had an average deviation of 0.6 octane numbers and a maximum deviation of 1 octane number.
To take maximum advantage of the benefits of the new system, it was necessary to consider the "integrated" value of several measurements, which is more representative of the tank and more accurate. For example, carrying out N measurements during the operation (every 45 sec) reduces the deviation by a factor of checkN.
Moreover, in terms of flow rate and basestock percentage, the difference between the blending order and the actual blend composition is very small (
The Anamel system works like a process-control system, using the NIR blending indices as the correlation matrix. The consistency of these indices therefore has to be great to avoid any drift of the blending control. Such drift has never happened at Lavera.
The NIR indices now are used for linear-programming scheduling and also for monthly evaluation of refinery performance.
The blender reliability, based on NIR analyzer performance, is 100%. This means that the right target is always reached from NIR information. In terms of real use, the reliability is 90%, the 10% default being caused by off site alarms involving computers, pumps, information-transfer links, etc.
Another indication of the system's success is that, since installation, reblending has never been required.
The CFR engines now are used only to check periodically the status of the NIR analyzers--both the laboratory and on-line systems. The frequency of motor-measurement comparisons against NIR analysis is shown in Table 2 for basestocks and finished products.
For the basestocks, no motor measurements, neither RON nor MON, have been performed, while 65 NIR analyses have been carried out. For the three finished product grades, 45 NIR analyses have been performed, vs. 10 engine measurements (5 RON, 5 MON). These engine tests have been performed within the time frame of maintenance procedures for the off-line and on-line NIR systems.
The values in Table 2 can be considered typical for a month of operation.
BENEFITS
The Lavera on-line NIR blending system offers many advantages. In terms of blending indices, all indices are derived from a single measurement. It is therefore possible to assess, at the same time, the suitability of a basestock for several gasoline grades. The oxygenate contribution is now determined easily and accurately, whatever the pool. This accuracy in determining blending indices permits the computation of more-precise blending orders.
The main result of this major improvement in the accuracy of calculating blending orders is the minimal deviation between finished-product measurements and specifications. This key point leads to a strong reduction of quality give-away, which is possible because lower internal specifications can be set for key properties.
For example, the internal MON specification was reduced by 0.3 points. The constraint cost for 1 octane number being $5/cu m ($0.79/bbl), the benefit of this reduction is about $150,000/month for a volume production of 100,000 cu m (about 21,000 b/d). The annual benefit thus is about $2 million, which ensures rapid pay-back for the system.
The analyses performed by the NIR instruments replace a number of classical analyses, such as knock-engine tests, and measurements of density, Rvp, distillations, and gums. This benefit is estimated to be about $300,000/year.
Additional advantages are, for instance, the integration of anticipated seasonal product specifications into the models, and an original approach to taking into account lead susceptibility response when calculating octane number. Because of this second benefit, the value determined by the model for a fixed lead level (say, 0.15 g/l.) can be corrected with the true value determined by classical analysis (for example, 0. 13 g/l.). The corrected results obtained from the model correlate perfectly with the CFR measurements performed on standard samples. BPs NIR technology is available through Technip.
REFERENCES
- Annual Book of ASTM Standards, Methods D2699-88a and D270088a.
- Swarin, S.J., and Drumm, C.A., SAE Technical Paper Series, Toronto, October 1991.
- Lambert, D., and Martens, A., Patent FR 2 611 911, 1987; EP 285 251, 1988.
- Lambert, D., Martens, A., and Ventron, G., Patents EP 304 232 and EP 305 090, 1988.
