- : Ms Word, Ms Word Format
- : 100 Pages
- : ₦5000
- : 1-5 Chapters
DETERMINATION OF A PROCEDURE FOR MONITORING SACRIFICIAL CATHODIC PROTECTION SYSTEMS FOR ABOVEGROUND STEEL STORAGE TANKS
Abstract
The main objective of this research was to develop a procedure for monitoring sacrificial cathodic protection systems that have been installed under steel aboveground storage tanks. The procedure was to include a method that would account for the IR drops, or potential losses, in the system. Once the losses are accounted for, a direct comparison to the required –850mV between the tank and the copper – to – copper sulfate (Cu/CuSO4) reference electrode can be made. This comparison will determine if the systems are providing the required protection for the storage tank.
The first step in accomplishing this goal was to formulate a list of variables that are thought to affect the potential losses in the system. Once completed, the task of building an experimental setup that facilitated the measurement of each of these variables became important. Many data sets, 288 specifically, were recorded using this controlled test setup. Measurements were taken during all seasons of the year and during all types of weather and temperatures, mirroring real world conditions.
Field data was also obtained using existing cathodic protection systems located in facilities across Pennsylvania and Ohio. Once all the data had been collected, appropriate files were created for use in a neural networking program. Eighty percent of the experimental data was used to train the system while the remaining twenty percent was used to test the system. After the final weights had been calculated by the network, the field data was tested in the network. The output became the adjusted voltage which could be directly compared to the required –850mV.
The back propagation neural network chosen for this project proved to be very useful. The network outputs of the training and testing phases were all within 4% of the actual outputs measured using the experimental tank. When the field data was tested with the network most of the systems were providing more protection than required which confirms the belief that the cathodic systems are indeed working as intended.
Table of Contents
List of Tables……………………………………………………………………………………………………vii List of Figures …………………………………………………………………………………………………viii List of Abbreviations …………………………………………………………………………………………. x Acknowledgements …………………………………………………………………………………………..xi
Chapter 1. Introduction and Background ………………………………………………………….. 1 Background on Cathodic Protection……………………………………………………………………. 2 Problem Statement……………………………………………………………………………………………. 7 Scope of Research ………………………………………………………………………………………….. 11 Background on Parallel Distributed Processing………………………………………………….. 14 Objectives……………………………………………………………………………………………………… 20
Chapter 2. Literature Review…………………………………………………………………………… 22 Cathodic Protection ………………………………………………………………………………………… 22 General Cathodic Protection System Design ……………………………………………………… 29 Parallel Distributed Processing…………………………………………………………………………. 30 Chapter 3. Research Approach ……………………………………………………………………….. 33
Chapter 4. Data Collection and Processing……………………………………………………….. 55 Data Collection………………………………………………………………………………………………. 55 Data Processing ……………………………………………………………………………………………… 60
Chapter 5. Parallel Distributed Processing……………………………………………………….. 64 Program Background………………………………………………………………………………………. 64 Program Implementation…………………………………………………………………………………. 67
Chapter 6. Results and Discussion …………………………………………………………………… 75 Experimental Data …………………………………………………………………………………………. 76 Experimental Data Results ………………………………………………………………………………. 86 Field Data Results…………………………………………………………………………………………. 101
Procedure for Monitoring Sacrificial Cathodic Protection Systems …………………….. 119
Chapter 7. Conclusions Recommendations for Further Study …………………………. 121
Conclusions …………………………………………………………………………………………………. 121 Recommendations for Further Study……………………………………………………………….. 123 Appendix A. Experimental Tank Data ……………………………………………………………. 125 Appendix B. Field Tank Data…………………………………………………………………………. 162
Appendix C. PDP Setup Files ………………………………………………………………………… 174 Template File……………………………………………………………………………………………….. 175
String File……………………………………………………………………………………………………. 177
Network File………………………………………………………………………………………………… 177 Training Patterns File ……………………………………………………………………………………. 179 Testing Patterns File……………………………………………………………………………………… 185 Field Patterns File…………………………………………………………………………………………. 187 Final Weights File ………………………………………………………………………………………… 191 References……………………………………………………………………………………………………… 192
Chapter 1 Introduction and Background
The main objective of this research was to develop a procedure for monitoring sacrificial cathodic protection systems that have been installed under steel aboveground storage tanks. The procedure was to include a method that would account for the IR drops, or potential losses, in the system. Once the losses are accounted for, a direct comparison to the required –850mV between the tank and the copper – to – copper sulfate (Cu/CuSO4) reference electrode can be made. This comparison will determine if the systems are providing the required protection for the storage tank.
The first step in accomplishing this goal was to formulate a list of variables that are thought to affect the potential losses in the system. Once completed, the task of building an experimental setup that facilitated the measurement of each of these variables became important. Many data sets, 288 specifically, were recorded using this controlled test setup. Measurements were taken during all seasons of the year and during all types of weather and temperatures, mirroring real world conditions.
Corrosion is a naturally occurring phenomenon that causes damage to almost everything from automobiles and home appliances, to bridges, pipelines, and public buildings. The damage caused by corrosion can be both dangerous and very expensive to rectify. According to the most recent study conducted between 1999 and 2001 by CC Technologies Laboratories, Inc. in conjunction with the National Association of Corrosion Engineers (NACE), the United States spends approximately $276 billion a year as a direct result of metallic corrosion. The direct cost for aboveground storage tanks alone is estimated to be $4.5 billion. It is believed that if optimum corrosion management practices were utilized, 25 to 30% of this cost could be eliminated (Koch 2002). Cathodic protection is a corrosion management technique. It is imperative to understand how well the preventative measures, such as cathodic protection, are working in order to lower the cost as much as possible.
Background on Cathodic Protection
To understand why corrosion is a problem and what the need is for cathodic protection it will first be helpful to explore what corrosion is and how it is affected by cathodic protection. Corrosion can be defined as the destruction or deterioration of a material because of reactions with its environment (Fontana, 1986). Corrosion can affect both metals & non-metals. For this research, discussion will be limited to corrosion of metals, specifically carbon steel aboveground storage tanks. Corrosion can be further divided into eight forms: (1) uniform, or general attack; (2) galvanic, or two-metal corrosion; (3) crevice; (4) pitting; (5) intergranular; (6) selective leaching or parting; (7) erosion corrosion; (8) stress corrosion. Some forms are unique but may share common characteristics. This research is concerned only with the electrochemical type of corrosion. This is the type of corrosion that aboveground steel storage tanks are being protected against with the sacrificial cathodic protection systems in question. There are four main components in a corrosion cell: anode, cathode, a metallic path connecting the anode and cathode, and an electrolyte (API, 1991). There are two main reactions that occur in a corrosion process: oxidation and reduction. Oxidation, which is a loss of electrons, occurs at the anode, and reduction, or electron gain, occurs at the cathode. A typical corrosion cell for an aboveground storage tank can be seen in Figure 1.1.
Figure 1.1 Typical Corrosion Cell
One of the most common ways of protecting metal against corrosion is cathodic protection. The definition of cathodic protection according to the National Association of Corrosion Engineers (NACE, 1991) is the reduction of corrosion rate by shifting the corrosion potential of the electrode toward a less oxidizing potential by applying an external electromotive force. There are two different types of cathodic protection systems; impressed current systems and sacrificial systems.
The aboveground storage tanks that are of interest to this research are unheated, carbon steel storage tanks. The type of steel used in the construction of tank bottoms is typically A36. The main element in steel is iron, and iron will corrode when in contact with any half-cell reaction higher on the electromotive force (EMF) series or even when just in contact with water. This contact produces a current that flows from the anode (the metal that is corroding) into the electrolyte. An impressed current system uses an external DC power source to mitigate corrosion by producing a current in the opposite direction that flows from the electrolyte into the anode. This causes the current to create a condition so that the electrons, which flow in the opposite direction of the current, flow into the anode and force the electrode potential of the iron from its open circuit corrosion potential in the negative direction to below irons half-cell equilibrium potential. The equilibrium potential, Eeq, is defined as the potential at which there is no net current through the electrochemical cell and the reaction is reversible, or proceeds equally fast in both directions. The corrosion potential, Ecorr, can be defined in terms of “polarization” which refers to a shift in potential caused by current flow. As more current flows from the anode into the electrolyte, the anode’s potential increases whereas the cathode’s potential decreases. Both electrodes will polarize until they reach essentially the same potential. This is the corrosion potential.
When two half-cell reactions from the EMF series are connected, the one that is the lowest in the series will be the one that is oxidized. Sacrificial cathodic systems connect the steel bottom, which is predominantly iron, to a metal that is lower on the EMF series, typically zinc or magnesium, utilizing the principle of galvanic corrosion to induce the desired protection. The zinc or magnesium becomes the anode and will sacrifice itself by oxidizing; thereby lowering the potential below Ecorr (partial protection) and even below Eeq for complete protection of the steel bottom. This procedure can be seen in Figure 1.2 which shows the polarization curves of zinc and iron when in contact with each other. Mixed potential theory, which is a procedure that is used to determine the corrosion condition (Ecorr) for systems which have multiple anodic and cathodic reactions by obtaining a sum of the respective branches to produce total anodic and cathodic polarization curves, is used to determine the Ecorr of the zinc-iron system. This research is only concerned about the sacrificial type of cathodic protection system. Figure 1.3 shows what a corrosion cell on a storage tank bottom would look like once cathodic protection were installed using zinc anodes.
Figure 1.2 Zn – Fe Galvanic System
Figure 1.3 Cathodic Protection Corrosion Cell
One of the important variables in cathodic protection is the voltage drop which is due to the relationship between current and resistance. It is customary to refer to the resistance-current relationship as an IR drop. These IR drops account for the voltage potential differences, resulting in varying degrees of protection at different locations along the surface.
When monitoring a cathodic protection system, potential measurements are taken between the protected structure and a reference electrode. A reference electrode is simply a piece of metal immersed in a solution of one of its salts (Ansuini and Dimond, 1994). If the reference electrode is thermodynamically stable, a known reversible chemical reaction will occur between the metal and its environment. If the reaction is at equilibrium, the rate is equal in both directions and will follow the Nernst equation as shown (Pickering, 2003):
E = E0 – (0.059/n) * log Q (1.1)
| where | ||
| E | = | equilibrium potential |
| E0 | = | standard potential |
| Q | = | ([C] c[D] d)/([A] a[B] b), given a general reaction of the form |
aA + bB+ ne– = cC + dD (1.2)
Theoretically, only two factors should affect a reference electrode’s potential:
temperature and solution concentration. There are two types of reference electrodes. The first is a metal in a solution containing dissolved ions of that metal such as the copper/copper sulfate electrode. The second is a metal coated with a salt of that metal and immersed in a solution of that salt such as the calomel and silver/silver chloride electrodes (Ansuini and Dimond, 1994). Reference electrodes should have constant potentials that current flow will not disturb, have low temperature coefficients, regain equilibrium quickly when mechanically disturbed, and be sturdy and durable (Pickering, 2003). A copper/copper sulfate (Cu/CuSO4) reference electrode will be used for this research. This type of electrode is commonly used for field measurements due to its robustness.
Problem Statement
When the bottom steel of a carbon steel storage tank is thinner than 0.1”, the
American Petroleum Institute (API) 653 code requires that the bottom be replaced or the tank be removed from service. Replacing a bottom is a very costly undertaking. On average a new bottom will cost about $1500 times the diameter of the tank to replace, approximately $180,000 for a 120’ diameter tank. One way to increase the life of a storage tank bottom is to utilize cathodic protection either under the bottom when a new tank is built or in the interstitial space between the old bottom and the new when replacing the bottom. Having a clear understanding of how the cathodic system is performing becomes very important. If the system is not performing as it should, a tank owner wants to know as soon as possible in order to fix the problem before the tank bottom starts to corrode, shortening the life of the tank.
The need for this research was recognized when evaluating field surveys of sacrificial cathodic protection systems. The direct voltage measurements taken in the field would suggest that the required protection is not being provided; however, anecdotal evidence indicates this is not accurate. When monitoring a sacrificial cathodic protection system, a minimum value of –0.85 volts between the Cu/CuSO4 reference electrode and the steel is the value required by the National Association of Corrosion Engineers (NACE) and by the American Petroleum Institute (API) to indicate adequate protection of the steel. To put the value of –0.85 volts in prospective, it can be compared to the equilibrium potential, Eeq, of iron which is –0.758 volts related to the copper sulfate reference electrode. This value of –0.85 volts is more negative than Eeq which theoretically completely protects the steel against corrosion. The required protection voltage of –0.85 volts should be the value at the soil-to-steel interface. However, voltage measurements in the field are taken utilizing a reference electrode that is permanently buried in the electrolyte several inches from the steel surface, or at the periphery of the tank. Therefore, in the case of cathodic systems in aboveground storage tanks, this value usually cannot be obtained (based on numerous field surveys). It is theorized that the cause of this discrepancy is due to certain factors such as resistance in the wires and in the electrolyte (wet sand), moisture and air content of the sand, the distance between the reference electrode and the soil-to-steel interface, and the applied pressure on the soil-tosteel interface. The resistance in the electrolyte is typically measured in ohms/cm versus the resistance in the wire that is typically measured in ohms/1000 ft; therefore, because the resistance in the wire is so much smaller than that of the electrolyte it can be neglected.
The applied pressure on the sand, the resistance, moisture and air content of the sand are all important influences on the voltage reading. When a tank bottom is constructed, the process of welding the bottom plates together causes geometric variations in the bottom as a whole and can cause portions of the steel to not be in complete contact with the supporting sand base. If there is enough product in the tank, the head pressure of the liquid will put the bottom into more uniform contact with the electrolyte (sand). When the product stored in the aboveground tank is at a low level, the weight applied to the bottom may not be enough to flatten it and to cause complete contact with the electrolyte (sand). If the bottom is not in complete contact there will be a higher air content in the electrolyte. This may cause the system to experience a higher IR drop due to the dielectric properties of air which: 1) give it a high resistance and 2) allows no current movement for low voltage systems such as this. The resistance of the electrolyte varies with the moisture content of the sand. Moisture content is inversely proportional to the resistance and therefore to the IR drop, which implies the higher the moisture content, the smaller the IR drop. Pressure on the sand is also related to the IR drop. The flatter the bottom, the more surface area that is available for current flow. Also, a flatter bottom leads to less air in the interstitial space and therefore less resistance and a smaller IR drop. Due to the resistance of the electrolyte, the IR drop between the reference electrode and the steel is proportional to the distance between them. The relevant equation (Morgan, 1993) is:
R = ρ(l/a2) (1.5)
where
R = Resistance in ohms
ρ = the resistivity of the electrolyte in ohm-cm a = width and thickness of the reference cell in cm assuming
approximately a rectangular shape
l = the distance between the reference cell and the tank bottom in cm. The assumption for this equation is that there is an infinite medium – an untrue assumption for this research; therefore, a sub-study will be conducted to determine if this assumption may be reasonably used for these cathodic protection systems.
Scope of Research
There are two parts to this study. The first examined each of the variables shown in Table 1.1 to determine the relationship between them and the voltage reading. Several of these variables cannot be measured for cathodic protection systems that have been installed under storage tanks in the field; therefore, an experimental test setup was constructed with extra monitoring devices included that allowed these variables to be measured. Data was collected from both the experimental test tank and tanks in the field for comparison. The relationship between these variables and the IR drop in the system is unknown and may be either linear or nonlinear. This type of data where a discreet solution is not easily accomplished by normal mathematical procedures is ideal for systems such as a parallel distributed processing system. These types of systems can
“learn” and adapt the solution by adjusting weights as needed to obtain a best fit solution.
| Table 1.1 List of Variables | |
| Variable # | Description |
| 1 | Resistance in reference electrode backfill * |
| 2 | Resistance in electrolyte * |
| 3 | Moisture content of electrolyte * |
| 4 | Pressure on soil/steel interface (fill height) |
| 5 | Distance to reference cell from the center of tank |
| 6 | Distance between reference cell and tank bottom |
| * Variables that cannot be measured for existing cathodic protection systems | |
The second phase of this project used a parallel distributed processing (PDP) computer program, or neural networking program to evaluate the data obtained both experimentally and in the field. Information in the program is represented through a series of connections and weights. These types of systems are thought to imitate biological systems through mathematical methods. The PDP system operates on many of the same principles as multiple regression; however, due to back-propagation, which takes the calculated error and adjusts the weights layer by layer, and the logistic function, this system can represent nonlinear information that was previously confounded with error using linear methods of analysis. A least squares type of solution using error to revise the estimates is used as well as a gradient descent method, in which infinitesimal steps are taken, for the nonlinearity (Meley, 1995).
Essentially the PDP program was trained and tested with the experimental data. Once the program had “learned” then it was used with the field data. This steps taken were then used to develop a procedure to determine if cathodic protection systems are working using the present measuring techniques. Simply described, the experimental data was used to train and test the network and then the field data sets were used to validate the network. The field data was put into the trained network and the output was the voltage that had been adjusted by the program accounting for the potential drops in the system. This voltage shows whether or not the cathodic systems are affording the required amount of protection. See Figure 1.3 for a flowchart illustrating the various phases and steps required to accomplish this research.
Figure 1.4 Flowchart of Scope of Research
Background on Parallel Distributed Processing
The second phase in this project is to utilize the principles of parallel distributed processing. Some background information will be helpful in understanding how it works. This section will also explore the various advantages that make this type of processing ideal for use in this research project. Parallel distributed processing , or neural networks, is a type of methodology that was originally based on a conceptualization of the human brain. The human brain is basically a highly complex, nonlinear, and parallel computer. The human brain is capable of organizing its neurons to perform certain computation such as pattern recognition, perception, and motor control at a much faster rate than any computer in existence. This capability comes because of the adaptability of the brain and its ability to learn through experience. Essentially a neural network is designed to perform tasks the way the brain does, through a learning process. A neural network is a massive parallel distributed processor made up of simple processing units, which has a natural propensity for storing experiential knowledge and making it available for use. It resembles the brain in two respects: (1) knowledge is acquired by the network from its environment through a learning process, and (2) interneuron connection strengths, known as synaptic weights, are used to store the acquired knowledge (Haykin, 1999).
Neural networks offer a number of useful properties. First, an artificial neuron can be linear or nonlinear. The network itself is nonlinear if it is comprised of an interconnection of nonlinear neurons. This nonlinearity is distributed throughout the network and can be very useful if the problem to be solved is inherently nonlinear. Secondly, the network can learn. This is accomplished by modifying the synaptic weights by applying a set of training samples. Thirdly, these networks are very adaptable. They are able to adapt their weights to account for changes in the environment. These networks are also able to not only provide information about the patterns that should be selected, but they can also provide information about the confidence of the decision. Neural networks have the potential for robust computation. According to Jeng et al (2004), artificial neural networks are especially suited for modeling the behavior of ill-posed problems in which: (1) the behavior is so complicated that no rule or only ones based on oversimplified assumptions leading to limited application is available; (2) the description of the behavior has intrinsic uncertainty; or (3) a complete simulation is so time consuming or expensive that an accurate estimation is necessary. Thirumalaiah et al (1998) added several more advantages of neural networks: (4) they can be applied without prior knowledge of the process; (5) they are suitable for dynamic forecasting problems because the weights can be updated as new data is obtained; (6) a small amount of error in the input does not significantly affect the output because of the distributed processing; (7) less data storage is required because it is not necessary to keep all past data in the memory; and (8) they do not require any exogenous input other than the data set used to train the system.
The structure of a neuron in the neural network is directly linked with the type of learning algorithm chosen. There are several types of algorithms that are commonly used and each of these algorithms can be one of three different types; single-layer feed forward, multilayer feed forward, and recurrent. The single-layer network has only an input layer and an output layer. It is called a single-layer network because computation takes place only in the output layer.
In a multilayer network, which is the type that will be implemented in this research, there are one or more hidden layers. The computation nodes in the hidden layer(s) are called hidden units. These units serve to act as a sort of liaison between the external input and the network output. Adding these hidden layers allows the network to utilize higher-order statistics. The network can be either fully or partially connected. A fully connected network is one in which every node in each layer is connected to every node in the adjacent forward layer. A partially connected network refers to a network that is missing some of those connections. Figure 1.3 shows the fully connected, multilayer network with one layer of hidden units that will be utilized in this research.
Figure 1.5 Multilayer PDP Network
The third type is a recurrent neural network. This type is different than feed forward networks such as the other two types because it contains at least one feedback loop. This type of network can be with or without hidden neurons.
The ability to “learn” from its environment is what makes neural networks different from standard computer programs. “Learning” in the context of neural networks is defined as a process by which the free parameters of a neural network are adapted through a process of stimulation by the environment in which the network is embedded (Haykin, 1999). This “learning” occurs through a process of adjustments to its synaptic weights and bias levels which happens after each iteration and follows a sequence of events: (1) the network is stimulated by an environment, (2) the network undergoes changes in its free parameters (synaptic weights and bias levels), and (3) the network responds in a new way to the environment because of the changes in its internal structure (Haykin, 1999). This is where algorithms enter the picture. Each algorithm makes the adjustment to the synaptic weights differently.
There are several different algorithms that can be used. This research utilizes the back propagation algorithm; therefore, it is the only one that will be described here. Back propagation learning has two passes through the layers of the network, a forward and a backward pass. In the forward pass, the weights in the network are fixed and a function signal moves forward through the different layers of the network and emerges at the output nodes. It is during the backward pass that the weights are adjusted based on an error-correction rule. An error signal is produced at the output node by subtracting the networks response from the desired, or target, response. This error signal is propagated backwards through the system and the weights are adjusted to make the networks response statistically closer to the target.
Back propagation has two properties that are responsible for both its advantages and disadvantages. The first is that it is simple to compute locally and secondly it performs stochastic gradient descent in weight space (Haykin, 1999). Back propagation is subject to what is referred to as the locality constraint which means that the computation performed by the neuron is only influenced by the neurons that are directly connected to it. This property is generally desired for several reasons: (1) artificial neural networks that perform in this manner are considered a metaphor for biological neural networks, (2) it provides a basis for a fault-tolerant network design, and (3) it favors the use of parallel structures as an efficient method for using artificial networks. The first point is actually disputed for such reasons as: the synaptic connections in an artificial network may assume weights are either excitatory or inhibitory which is unrealistic because in the actual nervous system neurons are one or the other, in artificial systems, hormonal and other types of global communications are ignored but in the actual system these types of information are very important for state-setting functions, and it also appears unlikely that the brain transmits information backward along the axons. However, the back propagation technique has been widely successful even in the field of neurobiological phenomena.
Another advantage of the back-propagation multilayer network can be seen in the field of function approximation. It has been shown that this type of system can approximate functions such as piecewise differentiable functions that are not differentiable in the traditional sense but only in the generalized sense. Back-propagation systems are also considered robust, computationally efficient, and useful in sensitivity analyses. There are however still several disadvantages. Back-propagation is stochastic in nature which means that it takes a meandering path to the minimum and therefore it converges slowly; however, it still may be the preferred method considering higher-order methods may not converge any quicker and will require more effort. Another possible problem is that the system may become trapped in a local minima, though several researchers such as Rumelhart and McClelland (1987) state that this problem rarely occurs in back-propagation networks especially with a larger number of hidden units being utilized.
Taking the advantages and disadvantages into account it is still felt that a multilayer system in conjunction with the back-propagation algorithm is the best choice of system for use with this research. This research utilizes the parallel distributed processing program written by Rumelhart and McClelland in 1987. The layout of the system including the number of input, hidden, and output units will be described in
Chapter Five along with the various equations that the program used to train the network.
Objectives
Now that the problem has been presented, it is necessary to understand the steps that must be taken to accomplish the objective of this research. The objective is to develop a procedure for monitoring sacrificial cathodic protection systems in unheated, vertical, carbon steel aboveground storage tanks using existing technology (i.e. reference electrodes) that will allow the voltage drops to be accounted for so that it can be decisively determined whether a tank is adequately protected against corrosion.
This objective will be accomplished by following these four steps:
- Design an experimental test set up accounting for the following variables: moisture content of the electrolyte (sand), distance between the reference electrode and soil to steel interface, radial placement of the reference electrode, resistance of the electrolyte, resistance of the reference electrode backfill, and the applied pressure on the soil interface.
- Gather experimental data varying the pressure on the soil interface (height of the product) for each data set and repeating the procedure as the sand dries out.
- Use parallel distributed processing to determine the relationship between all of the variables and the voltage reading using the experimental values recently obtained to train and test the network. Then the trained network will be used with the field measurements to determine whether the cathodic protection systems are providing adequate protection.
Use the data obtained from steps 1-3 to develop a procedure for monitoring cathodic protection systems used to protect aboveground storage tanks.
