BR112018004602B1 - METHOD FOR GENERATING A DIFFUSE LOGIC KNOWLEDGE BASE FOR A PROGRAMMABLE DIFFUSE LOGIC CONTROLLER, NON-TRAINER COMPUTER READABLE MEDIA AND PROGRAMMABLE DIFFUSE LOGIC CONTROLLER FOR CONTROLING A PROCESS - Google Patents

Abstract

METHOD FOR GENERATING A DIFFUSE LOGIC KNOWLEDGE BASE FOR A PROGRAMMABLE DIFFUSE LOGIC CONTROLLER, A NON-TRANSIENT COMPUTER READABLE MEDIUM AND A PROGRAMMABLE DIFFUSE LOGIC CONTROLLER FOR CONTROLING A PROCESS. The present invention relates to a method for generating the knowledge base used for a programmable fuzzy logic controller comprising the steps of determining the relevant input and output variables to be controlled; create artificial potential fields for each of said variables; sampling each of said potential fields in order to generate fuzzy logic association functions; compiling said fuzzy logic association functions into fuzzy logic sets; and mapping fuzzy logic set inputs to output fuzzy logic sets through a rule base. Relevant input and output variables are including: minimum, maximum and equilibrium values; a weight of importance; a nonlinearity value; a control direction; and information whether said variable is an input or output variable. Further provided is a programmable fuzzy logic controller whose fuzzy logic knowledge base is obtained by the described method.

Description

History of the Invention Field of Invention.

[0001] The present invention relates to a programmable fuzzy logic controller, which uses an artificial potential field approach to automatically generate the knowledge base used in a fuzzy logic controller, by which thousands of rules can be generated from multiple inputs and outputs and the association functions that describe these inputs and outputs.

Description of the related Technique.

[0002] The artificial potential field method was originally developed as an alternative collision avoidance system for robotic manipulators. The manipulator moved in an abstract field of mathematically determined forces, where the position to be reached would be defined as an attractive pole, and the obstacles to be avoided would be defined as repulsive surfaces.

[0003] Fuzzy logic is a mathematical system that analyzes analog input values in terms of logical variables that consider continuous values between 0 and 1. These logical variables are generally mapped as sets of association functions known as fuzzy logic sets . The collection of fuzzy logic sets used throughout the system to interpret inputs and outputs is known as the fuzzy logic knowledge base.

[0004] The application of fuzzy logic theory to control systems is known as fuzzy logic control. Although fuzzy logic control is one of the approaches used to implement advanced processor controllers (e.g., for use in chemical process control), it is highly dependent on expert knowledge for its configuration rather than empirical or theoretical modeling. In particular, experts are called upon to provide the appropriate fuzzy logic sets that will form the basis of the fuzzy logic rule.

[0005] After strong initial interest in fuzzy logic controllers of the 1980s and 1990s, it fell out of fashion after the year 2000, and since then there has been little academic or commercial interest. The problem that caused this decline in interest was the obstacle posed by the need for manual development of the fuzzy logic rule base by a human expert. Commercial resources as well as academic interests were transferred to machine learning systems that were not subject to this requirement, such as Neural Networks, Bayesian Networks, and Support Vector Machines. Machine learning systems need less specific knowledge of the underlying process for their practical application, which makes them preferable to fuzzy logic control systems.

[0006] The task of developing member functions and fuzzy logic rules by fuzzy logic experts is empirical work and difficult to maintain. Changes in the ranges of input or output variables of the controlled process would require changes to the rule base or association functions, only a fuzzy logic specialist is capable of performing. This need has limited the use of fuzzy controllers in practical industrial applications because of the scarcity and high cost of these trained professionals.

[0007] Some controllers were developed that allowed the adaptation of pre-established association functions and the base of the fuzzy logic rule. Although these controllers reduced the need for fuzzy logic experts, they required database and/or online learning experts for their operation. These new experts were necessary because a controller that is obtained through data analysis must be based on information representative of the process being controlled. If the database is not representative, the controller may exhibit unwanted and even dangerous behavior due to incorrect correlations derived from the provided data.

[0008] In a further attempt to increase the adoption of fuzzy controllers and overcome the need for experts to configure and tune these systems, the International Electrotechnical Commission (IEC) proposed Rule 61131-7, which required the standardization of controller programming of fuzzy logic in the year 2000, This IEC61131-7 standard was adopted by some manufacturers of Programmable Logic Controllers (PLC) and Distributed Control Systems (DCS), hoping to create an interchangeable platform of fuzzy logic controllers, but failed to solve the need for maintenance, a specialist problem, which continued to cause a decline in the adoption of fuzzy logic controllers in the industry.

[0009] EP 0355716 defines a rule generation method for fuzzy inference with the main benefits in automating rule resolution and automatically tuning database association functions. The automation of fuzzy rule generation represents an important advance in the dissemination of fuzzy controllers, a challenge to which EP 0355 716 contributed by modeling a fixed number of membership functions for a pre-established database. A large number of linguistic variables contribute to better resolution of the fuzzy controller, a small and fixed value in the aforementioned patent. The invention proposed in this document, Learning Fuzzy Control eliminates the need for previous databases and automatically determines the number of linguistic variables and association functions with the aim of optimizing the resolution of the automatically generated fuzzy controller.

[0010] Document EP 2169484 proposes a fuzzy logic system, specialized in cement production, which provides robust indicators of the temperature profile of cement rotary kiln equipment. The main benefit was to treat the information robustly, determine which information was contradictory in a first analysis and provide reliable information to a model predictive control (MPC) controller. The proposed patent now, in addition to automatically generating the fuzzy system, eliminates the need to use an MPC controller. Learning Fuzzy Control moves from analyzing indicators to writing directly to the final control element.

[0011] The current technique falls into providing a practical method for automatically generating a fuzzy logic knowledge base controller.

Summary of the Invention

[0012] The controller according to the present invention is capable of interacting directly with the process to be controlled through A/D and D/A converters, and the user does not need to have any knowledge of fuzzy logic to generate fuzzy controllers with multiple inputs and multiple outputs (MIMO). To develop the knowledge base of the fuzzy controller, it is necessary to provide the programmable controller (PC) with analog inputs that contain, for each variable to be controlled: the minimum (real domain), maximum (real domain), equilibrium point (real domain). real), importance weight (from 0% to 100%), nonlinearity (real domain), control direction (-1 or 1 for inverse and direct action control, respectively) and whether the variable is an input or an output (-1 or 1).

[0013] In a simplified form, Learning Fuzzy Control is carried out in three phases:

[0014] (i) Creation of potential fields for each variable (reading or writing) of the PC: the minimum, maximum and non-linearity equilibrium points are parameters of a linear and differentiable function that generates a linear potential field (F[ x]) for each variable. The potential field should always have its minimum value F[x] = 0 for x = equilibrium point.

[0015] (ii) Sampling of potential fields and generation of association functions: each potential field is sampled with small steps (x + Δx) with respect to F[x]. For each sampling xj a membership function is generated, the area of which is proportional to F[x], that is, the larger F[x], the larger the area of the association function. Therefore, the closer xj is to the equilibrium point value, the smaller the value of F[x] will be, and consequently the area of the association functions will also be smaller. Many association functions with reduced areas that generate an area of attraction around the equilibrium point value, replicating the behavior of the potential field in the set of linguistic variables of the fuzzy logic system.

[0016] (iii) Creation of the controller with fuzzy logic knowledge base: after the end of the preliminary phase, there is a set of association functions for each variable PC; In this phase, the variables are separated into input and output groups, and the input/output mapping is carried out according to the control signal information. If the control direction is 1 (direct), the first association functions of the input variable are connected to the first association functions of the output variable; if the control direction is -1 (reverse), the first association functions of the input variable are connected to the last association functions of the output variable.

[0017] After phase (iii), Learning Fuzzy Control generated a complete fuzzy logic control system that can be realized by a controller adhering to the international standard IEC 61131-7.

Brief Description of the Drawings

[0018] The present invention has other advantages and characteristics that will be more readily evident from the detailed description of the invention and the attached claims, when taken together with the attached drawings, in which:

[0019] Figure 1 is an illustration of a component for expressing variables in parameters through analog inputs;

[0020] Figure 2 is an illustration of a component that forms the base signal of the potential field for sampling variables;

[0021] Figure 3 is an illustration of an example of different functions that follow the restrictions of the minimum and maximum equilibrium point;

[0022] Figure 4 is an illustration of a summary of Fvn sampling to generate the association function;

[0023] Figure 4A is an illustration of interactions for sampling the left and right sides of F(x);

[0024] Figure 4B is an illustration of the fuzzy logic set using the nonlinearity value < 0;

[0025] Figure 4C is an illustration of the fuzzy logic set using nonlinearity value = 0;

[0026] Figure 4D is an illustration of the fuzzy logic set using nonlinearity value > 0;

[0027] Figure 5 is an illustration of the formation of exemplary fuzzy, forward and reverse logic rules;

[0028] Figure 6 is an illustration of a machine for automatic generation of the fuzzy logic knowledge base controller.

[0029] Figure 7 is the curve generated for Boiler pressure with a Non-Linear parameter equal to 0.000065 for the specific modality described.

[0030] Figure 8 is the curve generated for Steam flow with a Non-Linear parameter equal to 0.001152 for the specific modality described.

[0031] Figure 9 is the curve generated for the Energy flow with a Non-Linear parameter equal to 0.033975 for the specific modality described.

[0032] Figure 10 is the boiler pressure fuzzy logic set for the specific embodiment described.

[0033] Figure 11 is the vapor flow fuzzy logic set for the specific embodiment described.

[0034] Figure 12 is the set of energy flow fuzzy logic for the specific embodiment described.

[0035] Figure 13 is the mapping of the fuzzy logic rules between boiler pressure and energy flow (reverse direction) for the specific embodiment described.

[0036] Figure 14 is the mapping of the fuzzy logic rules between the steam flow rate and the energy flow (direct direction) for the specific modality described.

[0037] Figures 15-20 are flowcharts of operation in accordance with the present invention for producing a fuzzy logic knowledge base.

[0038] Figures 21-23 are operation flowcharts of a fuzzy logic controller using a fuzzy logic knowledge base developed in accordance with Figures 15-20,

[0039] Figure 24 is a block diagram of a computer for performing the flowcharts of Figures 15-20,

[0040] Figure 25 is a block diagram of an integrated fuzzy logic controller to realize the flowcharts of Figures 21-23.

Detailed Description

[0041] Artificial potential fields were initially used as algorithms for robot locomotion. The first mention of this approach was used by Oussama Khatib in 1986. The initial objective of this approach was to generate behaviors in robots to make them capable of autonomous navigation, avoiding (repulsive) obstacles and causing them to be attracted to pre-established goals. .

[0042] The attraction/repulsion surface can be quite flexible, from a cone (in a three-dimensional attractor) to complex shapes with singularities. Generating attraction surfaces for larger sizes is also no problem. It is only necessary to increase the number of input variables.

[0043] Currently, artificial potential fields are used in computer games to govern the artificial intelligence movements of opponents. The latter are attracted to human players and repelled by dangers, such as fire or the human players' shooting range, for example.

[0044] The artificial potential field approach derives from a simple premise: avoiding places that present harm (repulsion) and nearby places that present benefits (attraction).

[0045] An industrial controller must keep the control variable as close to the set point as possible, thus reducing variability as much as possible.

[0046] Safe and robust industrial controllers change actuator values gradually and without jerks. The mathematics present to generate the potential fields guarantees that there will be no jumps (if there is no singularity in the equation), and it is guaranteed that there will be only one point of attraction (the set point in the case of the control variable).

[0047] The decision of the attraction point for the actuator values and the disturbance variables is not so obvious. There is no real value for these variables. However, equipment operators and process engineers are able to estimate values that, from experience, are maintained when the process remains stable. These values obtained through experiment are defined as attractive for both the manipulated variables and the disturbance variables.

[0048] The main challenge was deciding which advanced control approach would be the ideal basis for applying the inspiration described above.

[0049] The traditional Proportional Integral and Derivative (PID) control strategy can be seen as a simplification of the inspiration of potential fields. The set point is the attractor and the factors Kp, Kd, Ki map between the difference between the set point and the controlled variable (controller error).

[0050] PID simplification implies limitations to controller modeling, such as:

[0051] Symmetry: positive errors in relation to the set point have an action of the manipulated variable with the same module as negative errors;

[0052] Linearity: the relationship between the controller error and the manipulated variable will be the same, regardless of the set point or operational characteristic of the equipment, since the entire relationship depends on the constants Kp, Kd, Ki alone;

[0053] Being based on accumulators: controller errors are weighted by the constants Kp, Kd, Ki and added to determine the value of the manipulated variable, this feature prevents "instantaneous" actions of the automatic controller. Another deficiency are systems with long times between the action of the manipulated variable and the impact on the value of the controlled variable (dead time); they are very difficult to tune (find good values for Kp, Kd, Ki);

[0054] One input variable and one output variable: the standard PID evaluates only the control variable for its decision on the manipulated variable. There are known modifications to read two or more variables (forward PID), but after three input variables it is very difficult to tune a PID. For example, some processes need to monitor more than ten variables (one control variable and nine disturbance variables).

[0055] The PID control approach has critical limitations that make it unfeasible as a base control platform for applying the inspiration of artificial potential fields. Other multivariable approaches to industrial process control are: control by process model and control through specialized systems.

[0056] Model-based controllers begin from the premise that it is possible to develop a model (linear or non-linear) of the process from a database that represents all states of the process or by elaborating differential equations that govern the process . In this way, the operation will be defined by the controller's inferences in the modeled process. The main technologies used in control through process modeling are neural networks and Model Predictive Control (MPC). Neural networks use a non-linear black box model of the process, unrelated to real physics. Therefore, they are subject to a lack of robustness and low extrapolation capacity. MPCs use linear dynamic models with multiple inputs and multiple outputs. As industrial processes have strong nonlinearity, MPCs cannot capture the entire plant dynamics. Additionally, all model-based controls need to be readjusted when there are process changes, which increases system maintenance effort and costs.

[0057] Expert systems aim to minimize the control actions desired by operation and process teams. Therefore, what is modeled is the ideal behavior of the controller, rather than the process itself. Typically, automatic control systems that use the expert systems approach are developed manually, that is, the rule base that governs the controller's behavior is defined manually by an expert.

[0058] The emulation of a behavior-based control strategy begins from the premise that this strategy is not implemented at present, either due to limitation in the execution of control actions at the speed of an algorithm or due to the analysis and interpretation of a large number of sensors available. Since there is no database, there is no way to perform learning from data.

[0059] Expert systems must be able to emulate the behavior of the human expert. The approach through artificial potential fields presents a useful simplification, reducing the range of all possible expert behaviors to only those behaviors that aim to reduce the overall error with respect to the attractors.

[0060] By transforming all variables into potential fields, whether inputs or outputs, the Fuzzy controller simplified the problem of working the rule base to the point where it is possible to automatically generate the Fuzzy knowledge base. The only information required a priori is: the upper and lower limits (MinVn and MaxVn), the equilibrium point of all variables (MedVn) and the assessment of how non-linear the variable is (NonLinear). With just these four types of information, it is possible to generate an attractive size for the chosen breakeven point. You must inform whether the configured variable is a reading or writing variable (IN/OUT). The weight of the variable in evaluating the value of the output/writing variable (weight) is another configuration factor for the input variables (if the configured variable is for writing/output, the weight information will be disregarded). Information must be entered if the input variable has a direct or reverse relationship (Dir/Rev) with the output variable (if the variable is an output, this information will not be considered). VN is a bus that has the information MinVn, MaxVn, MedVn, NonLinear, IN/OUT, Analog In, Dir/Rev for each reading or writing variable in the system.

[0061] The application of the potential field approach when generating fuzzy logic rules allows advanced multi-variable controller systems to be developed in a matter of minutes or hours, rather than months or years.

[0062] In Figure 1, the Input Variables component 100 is described, responsible for capturing the configuration parameters of each variable and transferring, through the bar, to the next component. The respective function is simply to convert the analog inputs, make them available on the bar and communicate them to the next component, Previous Signal. The parameters will be used to define the Fvn functions for each linguistic variable Vn.

[0063] In Figure 2, the component responsible for generating the potential fields defined by the Fvn function is described. This function should be continuous, with only a minimum value between the intervals of the variable's lowest configured value and its respective highest value. The minimum value of the F function should be exactly the equilibrium point value of the selected linguistic variable.

[0064] Below is an example to better demonstrate the component's function:

[0065] The Previous Signal component 200 composes the wave function for each linguistic variable according to the parameters: minimum, maximum equilibrium point and degree of non-linearity. This wave function is sent to the next component to perform sampling (also known as quantization) with a view to generating the pertinent fuzzy logic functions for all linguistic variables of the fuzzy logic process controller. It is important to observe some characteristics: asymmetric between the left side of the speech universe in Figure 3 (between minimum and equilibrium point) and the right side (between equilibrium point and maximum). Another important factor is the degree of nonlinearity that can be imposed on Fvn(x) through different derivatives between the left and right sides of the Fvn(x) function.

[0066] The Signal Sampler component 400 represents the link between the artificial potential field approach and fuzzy logics. In the proposed example, association functions are presented that are formed as triangular functions with a maximum height value equal to 1. Other association functions, such as Gaussian (bell-shaped) can be used. All membership functions sampled from the Fvn function compile a set of fuzzy logic, as described in Figure 4.

[0067] An example of sampling is shown in Figure 4A. Two sampling efforts are carried out: one for the left side and the other for the right side. This is necessary because MedVn may be closer to MinVn or MaxVn. It is important to start and end each of the linguistic variables with association functions with a high value. Therefore, in the MinVn and MaxVn values there will be association functions with a value of 1.0 (in the case represented by the first and last right triangles). The size of the base of these triangles will be proportional to the Fvn (xi), where the base of the first right triangle will be proportional to the first sampling in the left Fvn (Min Vn + xi) and the base of the last right triangle will be proportional to the last sampling in the Fvn right (Med Vn + vi).

[0068] The interactions in Figure 4A are intended to illustrate the process of filling the other association functions to fill the entire range of the fuzzy set. When using the maximum of the membership function, the value 1.0 (default when normalizing fuzzy priors), and isosceles triangles, the area of each membership function will be equal to Fvn(xi)/2. This simplification aims to increase the speed of obtaining association functions.

[0069] Figures 4B-4D described below show the differences between the association functions for the configuration parameters of the proposed example, using different values for Fvn nonlinearity. Figure 4B shows the format of the association functions for nonlinearity of Fvn less than 0. It is possible to notice that the area of the triangles is quickly reduced as the MedVn value is approached. Around the MedVn value there are a large number of association functions, generating great granularity and high resolution around this region. The asymmetry between the left and right side causes the area of association functions to be reduced more quickly on the right side than on the left side.

[0070] It is possible to generate a linear distribution of association functions by dividing only the speech universe (fuzzy logic set range) by the desired number of association functions. The machine proposed in this innovation automatically decides the number of association functions in relation to the universe of speech, as shown in Figure 4C. The number of association functions follows the same model for nonlinearity <0 and nonlinearity>0. The latter aims to maintain a balance between the granularity required for high resolution in mapping the universe of speech and the time to resolve this linguistic variable. The number of association functions may be different for each linguistic variable, and the number of association functions on the left side of the speech universe (below the break-even value of that linguistic variable) may be different from the number on the right side.

[0071] As the nonlinearity value increases, the granularity increases at values far from the MedVn. This effect can be seen in Figure 4D. More dispersed granularity means a distribution of importance of all values in the speech universe (to the detriment of a high concentration around the MedVn value, as shown in Figure 4B).

[0072] After the automatic distribution of the association functions into fuzzy logic sets, with the desired granularity, observing the universe of speech of all variables, it is time to form the base rules of fuzzy logic knowledge coupled to the inputs of the controller at its outputs.

[0073] The previous component created all sets of fuzzy logic from the selected variables, as shown in Figures 4B, 4C and 4D. The first phase of the multiplexing component is to separate the reading variables from the writing variables (or input variables from the output variables). The information used for this separation was inserted in the Input Variables component 100, IN/OUT information. If its value is high, this will be an input to the controller and if its value is low, it will be an output.

[0074] After separating the controller variables into two groups of variables, input and output groups, this is the time to form the fuzzy logic rules that will make up the knowledge base of the fuzzy logic controller. For each reading variable, the DIR/VER information provided to the Input Variables component 100 will be evaluated. If the signal configured for DIR/VER is high, the direction between the writing variable and the reading variable will be direct, that is, low values for the input variable will imply lower values for the output variable. If the signal configured for DIR/REV is low, the direction between the writing variable and the reading variable will be reversed, that is, low values for the input variable will imply higher values for the output variable. Translating this information into the formation of fuzzy logic rules, linking between the association functions of the reading and writing variables means that, in the forward direction, the initial association functions of the reading variable should be linked to the initial association functions of the writing variables. On the other hand, in the reverse direction, the initial membership functions of the reading variable are linked to the final variables of the writing variable, as shown in Figure 5.

[0075] It is clear in Figure 5 that the order of the association functions of the reading variables in the forward direction (upper example) is reversed in relation to the example in the reverse direction (lower example).

[0076] The physical effect between forward or reverse distribution is when the value of a reading variable increases, the value of the writing variable increases (if the direction is direct). If the direction is reversed, when the value of a read variable increases, the value of the write variable decreases.

[0077] The output of the Multiplex component or IN/OUT Selector 500 represents a full fuzzy logic (CF) advanced process control strategy with multiple inputs and multiple outputs for simple user configuration.

[0078] The complete machine can be designed according to the illustration shown in Figure 6. As can be seen, there are only five analog and two digital configurations for each variable of the fuzzy logic controller. The subsequent components are responsible for handling configurations, creating association functions and relating them to fuzzy rules, thus generating a knowledge base usable by a standard fuzzy logic interpreter, such as FCL.

[0079] Current industrial processes produce considerably more information than is actually used by automatic process controllers. From now on, three practical applications are presented, where the use of Learning Fuzzy Control allowed the use of more information made available by the sensors, increasing the efficiency of the process.

[0080] Example one: an energy cogeneration system fed with steam from gas and biomass boilers. Traditional automatic controllers (PID) are limited to keeping the pressure of each individual boiler stable and maintaining the power generation load of the turbines. It depends on the manufacturing processes and fuel supply to the boilers, preventing a large drop in pressure. The control system applied through Learning Fuzzy Control monitors the following variables: pressure of each boiler, general steam pressure, temperature of each oven, process steam flow, steam flow generated by each boiler, the controller decides what should be the flow of energy to the boilers (gas or biomass). The greater number of variables monitored in relation to the traditional control strategy practically eliminated the need for operator intervention, because the control system anticipates disturbances and acts preventively on the fuel supply, in addition to reducing gas consumption by 6%, a as it favors the biomass boiler, which is more economical, to the detriment of gas boilers. The variation in vapor pressure has been reduced by 30%. If the customer installs a humidity sensor on the biomass boiler, the performance of the controller would be even higher. The client does not have a history of reading variables (pressure of each boiler, general steam pressure, temperature of each oven, process steam flow, steam flow generated by each boiler), which makes training a fuzzy controller through the data is unfeasible. A manually developed multivariable fuzzy logic controller must have at least five triangular membership functions for each read variable and one singleton variable for writing. Therefore, eight reading variables (four for each boiler) plus the global steam pressure generate nine reading variables. Considering the two read variables with 5 singletons each, they generate 450 (45 association functions multiplied by 10 singleton reads) for the complete mapping of the knowledge base. In addition to the intense work involved in tuning the 45 association functions and 450 weights (one for each rule), the small number of association functions increases the tuning challenge due to the low resolution of the linguistic variable. The increase in association functions contributes to increasing the robustness of the fuzzy logic controller.

[0081] Description for practical obtaining of the fuzzy logic controller for a boiler described in the first example:

[0082] 1) The reading variables (INPUTS) were: boiler pressure and steam flow. The only writing variable (OUTPUT) was the energy flow in the boiler (biomass or gas).

[0083] 2) For each variable it is necessary to define which operation is desired, informing the minimum value (MinVn), the maximum value (MaxVn) and the average value (MedVn). At this time, it also defines the direction between the reading and writing variables (Dir/Ver) and the weight (Weight) of each reading variable in the writing variables.

[0084] 2.1) The minimum value (MinVn) for the boiler pressure was set at 43 kgf/cm2, the maximum value at 46 kgf/cm2, the average value (MedVn), netse if the set point, at 44, 5 kgf/cm2 and the direction as reverse (Rev). The weight (Weight) proposed for the boiler pressure was 95%.

[0085] 2.2) The minimum value (MinVn) for each steam flow was defined at 65 t/h, the maximum value at 80 t/h, the average value (MedVn) at 75 t/h and the direction as direct (Dir ). The weight (Weight) proposed for the steam flow was 55%.

[0086] 2.3) The minimum value (MinVn) for the energy flow was set at 5%, the maximum value at 80%, the average value (MedVn) at 40%. It is not necessary to define the direction of the Outputs.

[0087] 3) The final adjustment values will now be presented, the non-linear values and the weights obtained when evaluating the behavior of the controller with the boiler. The FVn (x) curves inserted by the Previous Signal are:

[0088] 3.1) Boiler pressure with Non-Linear parameter is equal to 0.000065. The generated curve is shown in Figure 7.

[0089] 3.2) Steam flow with Non-Linear parameter is equal to 0.001152. The generated curve is shown in Figure 8.

[0090] 3.3) Energy flow with Non-Linear parameter is equal to 0.033975. The generated curve is shown in Figure 9.

[0091] 4) The curves presented in 3) are sampled to define the bases of the membership functions that will make up the fuzzy logic sets for each variable. The samples for each variable performed by the Signal Sampler are provided in Table 1.Table 1

[0092] 5) The values sampled in 4) are used to form the membership functions that make up the fuzzy logic sets performed by the Signal Sampler component:

[0093] 5.1) Boiler pressure fuzzy logic set is shown in Figure 10,

[0094] 5.2) Steam flow fuzzy logic set is shown in Figure 11.

[0095] 5.3) Power flow fuzzy logic set is shown in Figure 12.

[0096] 6) With the fuzzy logic sets defined in (4), the Control Multiplex component maps the INPUTS (boiler pressure and steam flow) to the OUTPUTS (energy flow). The boiler pressure has a reverse direction in relation to the energy flow, whereas the steam flow has a direct direction in relation to the energy flow.

[0097] 6.1) Mapping of the fuzzy logic rules between the boiler pressure and the energy flow (reverse direction) is shown in Figure 13.

[0098] 6.2) Mapping of fuzzy logic rules between steam flow and energy flow (direct direction) is shown in Figure 14.

[0099] 7) With the base of the fuzzy logic rule mapped in (6), the Learning Fuzzy Control equipment completed the process of automatically generating the fuzzy logic rules. The multivariable controller of the industrial processes proposed in this example is available for execution by a fuzzy interpretation module. An example of a module is the script interpreter proposed by the IEC61131-7 specification, which defines the Fuzzy Control Language (FCL). A possible controller mapping generated by Learning Fuzzy Control compatible with the IEC61131-7 specification is presented in Table 2 below.Table 2

[00100] Example two: application in aluminum digesters - The controlled variable (or process variable) is evaluated through laboratory tests, and its results are made available every two hours. This fact makes a traditional application with PID unfeasible. The aluminum content present in bauxite varies during the day, depending on where it was extracted. The control of these processes is the responsibility of the operators, that is, it is in manual operation. Learning Fuzzy Control allowed simultaneous evaluation of the online variables of paste flow, bauxite flow, caustic refrigerant density, bauxite flow and digester temperature, in addition to offline variables (updated every two hours by the laboratory), such as caustic content in the digester, dissolved alumina content, solids concentration, tank coolant alumina/caustic ratio to write the best slurry flow rate for the process at the time. All values mentioned are evaluated and considered in each control cycle lasting 1 second. The application of the controller turned the 100% manual process into a 100% automatic process, reduced the variability of the controlled variable by 50% and reduced steam consumption by 3.2%. If bauxite density sensors were implanted at the digester inlet, sensors for online alumina concentration, sensor for decanted alumina in the subsequent digester process, the performance of the controller would be increased. The nine reading variables and one writing variable of the described control, using the methodology for developing a manual controller presented in the previous example, would require the adjustment of 225 rules. An added challenge is that offline variables are updated every two hours, making timely tuning impractical.

[00101] Example three: ammonia absorption towers in a nickel plant - ammonia diluted in low concentration water needs to be recovered to return to the process. This liquid was contaminated with various metals, making the process and absorption difficult. The main challenge is that the operation of the absorption towers needs to be interrupted every 60 hours due to fouling. Chemical or physical cleaning must be carried out to make the tower operational again. The controller implemented in this equipment, in addition to monitoring the sensitive temperature of the equipment (like a traditional PID controller), monitors the upper temperature, lower temperature, upper pressure, lower pressure, feed flow, steam pressure installations and steam flow, on-line boiler rate, bottom ammonia contamination are monitored through laboratory data updated every 4 hours. The controller shows the diluted ammonia feed flow and vapor flow to the absorption tower. With more variables to be monitored, Learning Fuzzy Control provided savings of 3.5% in steam consumption by the process, in addition to reducing operator intervention in starting and stopping maneuvers of the absorption towers. The eight reading variables and two writing variables of the controller described, using the methodology for developing a manual controller presented in the first example, would require tuning of 400 rules. In practical application, Learning Fuzzy Control was implemented in six absorption towers, which would increase the number of rules to 2400. An effort that would require more than a year of specialist work was carried out by a technician in less than a month.

[00102] An operation flowchart in accordance with the present invention is shown in Figure 15. In step 1502 the variable parameters are collected from the user, as shown in Figure 16. In step 504 the variable curves are generated, as shown in Figure 17. In step 1506, membership functions are generated, as shown in Figure 18. The generated association function of V(n) from Figure 18 is shown in Figure 19. In step 1508, fuzzy logic rules are generated, as shown in Figure 20. In step 1510 the fuzzy logic controller, the developed fuzzy logic knowledge base rules as described above are stored as a fuzzy logic controller 1512.

[00103] Figure 21 is a flowchart of operations of an integrated fuzzy logic controller. In step 2102 the fuzzy logic controller or knowledge base 1512 is loaded. In step 2104 the analog values of the process being controlled are loaded, as shown in Figure 22. In step 2106 the fuzzy logic control is operated, as shown in Figure 23, which includes providing the desired value of the output variable. In step 2108, if the execution is not completed, the operation returns to step 2104 for continued process control.

[00104] A computer, such as computer 2402 shown in Figure 24, can be used to generate the fuzzy logic knowledge base in accordance with the present invention. The computer 2402 includes a processor 2404, a memory 2406 for holding programs and data while operating, storage 2408 for storing the programs used to operate the computer 2402, and the fuzzy logic knowledge base or controller 1510 and a device human input/output 2410 to receive the variable values.

[00105] A fuzzy logic controller 2502, as would be used in a process such as those described above is shown in Figure 25. An analog-to-digital converter 2504 provides the input signals to an integrated fuzzy logic controller 2506, a computer configured to receive process inputs, operate on those inputs using the fuzzy knowledge base, and provide the determined output. A digital-to-analog converter 2508 receives the output signal from the integrated fuzzy logic controller 2506 and converts the digital value to an analog value used in controlling the process. The integrated fuzzy logic controller 2506 includes a processor 2510 for receiving digital input signals and providing digital output signals, a memory 2512, and storage 2514, which includes the fuzzy logic knowledge base 1510 used to control the process. This is a highly simplified block diagram and other configurations are known to those skilled in the art who can control the process using the developed fuzzy logic knowledge base, such as controllers that can be programmed in accordance with Rule IEC 61131-7.

[00106] It is understood that the fuzzy logic controller 2502 could also be used to develop the fuzzy logic knowledge base if software as described above and a way of receiving the variable information is provided.

[00107] The above description is illustrative and not restrictive. Many variations of the invention will become apparent to those skilled in the art upon review of this description. Accordingly, the scope of the invention must be determined not with reference to the above description, but rather with reference to the appended claims together with their full scope of equivalents.

Claims (20)

1. Method for generating a fuzzy logic knowledge base for a programmable fuzzy logic controller used to control a process, characterized by the fact that it comprises the steps of: receiving values of relevant input variables and output variables to be controlled; creating artificial potential fields for each of said relevant input variables and output variables; sampling each of said potential fields in order to generate fuzzy logic association functions; compile fuzzy logic association functions into fuzzy logic sets; and mapping the fuzzy logic sets of relevant input variables to output variables to create fuzzy logic rules, said fuzzy logic rules forming the fuzzy logic knowledge base. 2. Method, according to claim 1, characterized by the fact that the step of determining the relevant input variables and output variables includes, for each variable, providing: minimum, maximum and equilibrium values; a weight of importance; a nonlinearity value; a direction of control; and information whether said variable is an input variable or output variable. 3. Method according to claim 1, characterized by the fact that the step of creating artificial potential fields includes, for each variable, deriving a function that generates potential fields based on said relevant input and output variables, in that the potential field has its own minimum value at an equilibrium value of each variable. 4. Method, according to claim 1, characterized by the fact that the step of sampling each of said potential fields includes the steps of: sampling each potential field in small increments of value; and generating a fuzzy logic association function for each small increment sampled, where the area of the association function is proportional to the value of the sampled potential field. 5. Method, according to claim 1, characterized by the fact that the step of compiling said fuzzy logic association functions into a fuzzy logic set includes the steps of: separating said variables into input and output groups; and group the membership functions into fuzzy logic sets for each variable. 6. Non-transitory computer-readable medium that stores instructions for execution by a processor, characterized in that the processor executes the instructions to perform a method generating a fuzzy logic knowledge base for a programmable fuzzy logic controller used to control a process, the method comprising the steps of: receiving values of relevant input variables and output variables to be controlled; creating artificial potential fields for each of said relevant input variables and output variables; sampling each of said potential fields in order to generate fuzzy logic association functions; compile fuzzy logic association functions into fuzzy logic sets; and mapping fuzzy logic sets of relevant input variables to output variables to create fuzzy logic rules, said fuzzy logic rules forming the fuzzy logic knowledge base. 7. Non-transitory computer-readable medium according to claim 6, characterized by the fact that the step of determining the relevant input variables and output variables includes, for each variable, providing: minimum, maximum and balance; a weight of importance; a nonlinearity value; a control direction; and information whether said variable is an input variable or output variable. 8. The non-transitory computer readable medium of claim 6, wherein the step of creating artificial potential fields includes, for each variable, deriving a function that generates a linear potential field based on said variables of relevant input and output, where the potential field has its own minimum value at an equilibrium value of each variable. 9. The non-transitory computer readable medium of claim 6, wherein the step of sampling each of said potential fields includes the steps of: sampling each potential field in small increments of value; and generating a fuzzy logic association function for each small increment sampled, where the area of the association function is proportional to the value of the sampled potential field. 10. The non-transitory computer readable medium of claim 6, wherein the step of compiling said fuzzy logic association functions into a fuzzy logic set includes the steps of: separating said variables into groups entry and exit; and group membership functions into sets of fuzzy logic for each variable. 11. Computer for generating a fuzzy logic knowledge base for a programmable fuzzy logic controller (2506) used to control a process, the computer comprising: a processor (2404); an input device (2410) coupled to said processor (2404) for receiving input from a user; memory (2406) coupled to said processor (2404); storage (2408) coupled to said processor (2404) that stores instructions for execution by said processor (2404), wherein said processor (2404) executes said instructions to perform a method generating a fuzzy logic knowledge base for a programmable fuzzy logic controller, the method comprising the steps of: receiving values of relevant input variables and output variables to be controlled; creating artificial potential fields for each of said relevant input variables and output variables; sampling each of said potential fields in order to generate fuzzy logic association functions; compile fuzzy logic association functions into fuzzy logic sets; and mapping fuzzy logic sets of relevant input variables into output variables to create fuzzy logic rules, said fuzzy logic rules forming the fuzzy logic knowledge base. 12. Computer, according to claim 11, characterized by the fact that the step of determining the relevant input variables and output variables includes, for each variable, providing: minimum, maximum and equilibrium values; a weight of importance; a nonlinearity value; a direction of control; and information whether said variable is an input variable or output variable. 13. Computer, according to claim 11, characterized by the fact that the step of creating artificial potential fields includes, for each variable, deriving a function that generates a linear potential field based on said input variables and relevant outputs, in which the potential field has its own minimum value at an equilibrium value of each variable. 14. Computer, according to claim 11, characterized by the fact that the step of sampling each of said potential fields includes the steps of: sampling each potential field in small increments of value; and generating a fuzzy logic association function for each small increment sampled, where the area of the association function is proportional to the value of the sampled potential field. 15. Computer, according to claim 11, characterized by the fact that the step of compiling said fuzzy logic association functions into a set of fuzzy logic includes the steps of: separating said variables into groups of entrance and exit; and group membership functions into sets of fuzzy logic for each variable. 16. Programmable fuzzy logic controller (2502) for controlling a process, characterized in that it comprises: an analog to digital converter (2504) for connection to the analog input signals of a process to be controlled; a digital to analog converter (2508) for providing analog output signals to control the process; a processor (2510) coupled to said analog-to-digital converter (2504) and said digital-to-analog converter (2508); an input device coupled to said processor (2510) for receiving input from a user; memory (2512) coupled to said processor (2510); storage (2514) coupled to said processor (2510) that stores instructions for execution by said processor (2510), wherein said processor (2510) executes said instructions to perform a method generating a fuzzy logic knowledge base for a programmable fuzzy logic controller and control a process using the fuzzy logic knowledge base, the method comprising the steps of: receiving values of relevant input variables and output variables to be controlled; creating artificial potential fields for each of said relevant input variables and output variables; sampling each of said potential fields in order to generate fuzzy logic association functions; compile fuzzy logic association functions into fuzzy logic sets; mapping fuzzy logic sets of relevant input variables to output variables to create fuzzy logic rules, said fuzzy logic rules forming the fuzzy logic knowledge base (1510); receiving input values of the process to be controlled from said analog-to-digital converter (2504); and operating the received input values using said fuzzy logic knowledge base (1510) to provide an output to said digital-to-analog converter (2508) for controlling the process. 17. Controller, according to claim 16, characterized by the fact that the step of determining the relevant input variables and output variables includes, for each variable, providing: minimum, maximum and equilibrium values; a weight of importance; a nonlinearity value; a control direction; and information whether said variable is an input variable or output variable. 18. Controller, according to claim 16, characterized by the fact that the step of creating artificial potential fields includes, for each variable, deriving a function that generates a linear potential field based on said input variables and relevant outputs, in which the potential field has its own minimum value at an equilibrium value of each variable. 19. Controller, according to claim 16, characterized by the fact that the step of sampling each of said potential fields includes the steps of: sampling each potential field in small increments of value; and generating a fuzzy logic association function for each small increment sampled, where the area of the association function is proportional to the value of the sampled potential field. 20. Controller, according to claim 16, characterized by the fact that the step of compiling said fuzzy logic association functions into a set of fuzzy logic includes the steps of: separating said variables into groups of entrance and exit; and group membership functions into sets of fuzzy logic for each variable.