DOUBLE PIPE HEAT EXCHANGER FOULING COMPENSATION

Abstract

This disclosure presents methods and systems of controlling a counter flow double pipe heat exchanger (DPHE) that includes a hot fluid pipe and a cold fluid pipe. In a method, a temperature error between a reference temperature and a temperature at an outlet of the hot fluid pipe of the counter flow DPHE is determined. A cold fluid mass flow rate is determined from an output of a proportional-integral-derivative (PID) controller based on the temperature error being input to the PID controller. The cold fluid mass flow rate is used for a cold fluid in the cold fluid pipe of the counter flow DPHE. The temperature error is controlled within a predefined range by utilizing parameters of the PID controller that are set by using a harmony search algorithm (HSA) to obtain a minimization of a cost function.

Claims

1. A method of controlling a counter flow double pipe heat exchanger (DPHE) that includes a hot fluid pipe and a cold fluid pipe, the method comprising: determining a temperature error between a reference temperature and a temperature at an outlet of the hot fluid pipe of the counter flow DPHE; determining a cold fluid mass flow rate from an output of a proportional-integral-derivative (PID) controller based on the temperature error being input to the PID controller, the cold fluid mass flow rate being used for a cold fluid in the cold fluid pipe of the counter flow DPHE; and controlling the temperature error within a predefined range by utilizing parameters of the PID controller that are set by using a harmony search algorithm (HSA) to obtain a minimization of a cost function.

2. The method of claim 1, wherein the hot fluid pipe and the cold fluid pipe are an inner pipe and an outer pipe of the counter flow DPHE, respectively.

3. The method of claim 1, wherein the parameters of the PID controller include a proportional gain K.sub.p, a derivative gain K.sub.d, and an integral gain K.sub.i.

4. The method of claim 3, wherein 0≤K.sub.p≤5, 0≤K.sub.d≤10, and 0≤K.sub.i≤10.

5. The method of claim 1, wherein the cost function is given by $\mathrm{ost}\mathrm{function}=\frac{1}{N}{.Math.}_{k=1}^{N}k.Math.ϵ\left(k\right).Math.+{10}^3*\sqrt{\frac{1}{N-1}{.Math.}_{k=1}^N{\left(u\left(k\right)-u\right)}^2},$ wherein N denotes a total number of time samples during the minimization of the cost function, k denotes a time index, ϵ denotes the temperature error, u∈[0,2] denotes the output of the PID controller, and μ denotes a mean value of the outputs of the PID controller during the minimization of the cost function.

6. The method of claim 5, wherein the minimization of the cost function includes multiple iterations, wherein each iteration includes: generating a first plurality of candidate parameters from a second plurality of parameters according to a pitch adjusting rate (PAR) and a harmony memory considering rate (HMCR) of the HSA, the second plurality of parameters being stored in a harmony memory (HM); determining whether a result of the cost function based on a subset of the first plurality of candidate parameters is less than a result of the cost function based on a subset of the second plurality of parameters; in response to a determination that the result of the cost function based on the subset of the first plurality of candidate parameters is less than the result of the cost function based on the subset of the second plurality of parameters, replacing the subset of the second plurality of parameters with the subset of the first plurality of candidate parameters in the HM; and in response to a determination that the result of the cost function based on the subset of the first plurality of candidate parameters is not less than the result of the cost function based on the subset of the second plurality of parameters, removing the subset of the first plurality of candidate parameters.

7. The method of claim 6, wherein the PAR is given by x.sub.new=x.sub.old+FW.Math.ϵ, wherein x.sub.old denotes a current pitch, FW is a fret width, ϵ∈(0,1] is a random number, and x.sub.new denotes an adjusted pitch.

8. The method of claim 7, wherein $FW\left(i\right)=F{W}_{\max }\exp \left(\ln \left(\frac{F{W}_{\min }}{F{W}_{\max }}\right)*\frac{i}{\mathrm{NI}}\right),$ wherein FW.sub.max denotes a maximum fret width, FW.sub.min denotes a minimum fret width, NI is a total number of iterations, and i is a current iteration number.

9. A heat exchanging system, comprising: a counter flow double pipe heat exchanger (DPHE) configured to perform a heat exchanging process between a hot fluid in a hot fluid pipe and a cold fluid in a cold fluid pipe; and a proportional-integral-derivative (PID) controller configured to: receive a temperature error between a reference temperature and a temperature at an outlet of the hot fluid pipe of the counter flow DPHE, and output a cold fluid mass flow rate for the cold fluid in the cold fluid pipe of the counter flow DPHE based on the received temperature error, wherein the temperature error is controller within a predefined range by utilizing parameters of the PID controller that are set by using a harmony search algorithm (HSA) to obtain a minimization of a cost function.

10. The heat exchanging system of claim 9, wherein the hot fluid pipe and the cold fluid pipe are an inner pipe and an outer pipe of the counter flow DPHE, respectively.

11. The heat exchanging system of claim 9, wherein the parameters of the PID controller include a proportional gain K.sub.p, a derivative gain K.sub.d, and an integral gain K.sub.i.

12. The heat exchanging system of claim 11, wherein 0≤K.sub.p≤5, 0≤K.sub.d≤10, and 0≤K.sub.i≤10.

13. The heat exchanging system of claim 9, wherein the cost function is given by $\mathrm{ost}\mathrm{function}=\frac{1}{N}{.Math.}_{k=1}^{N}k.Math.ϵ\left(k\right).Math.+{10}^3*\sqrt{\frac{1}{N-1}{.Math.}_{k=1}^N{\left(u\left(k\right)-u\right)}^2},$ wherein N denotes a total number of time samples during the minimization of the cost function, k denotes a time index, ϵ denotes the temperature error, u∈[0,2] denotes the output of the PID controller, and μ denotes a mean value of the outputs of the PID controller during the minimization of the cost function.

14. The heat exchanging system of claim 13, wherein the minimization of the cost function includes multiple iterations, wherein each iteration includes: generating a first plurality of candidate parameters from a second plurality of parameters according to a pitch adjusting rate (PAR) and a harmony memory considering rate (HMCR) of the HSA, the second plurality of parameters being stored in a harmony memory (HM); determining whether a result of the cost function based on a subset of the first plurality of candidate parameters is less than a result of the cost function based on a subset of the second plurality of parameters; in response to a determination that the result of the cost function based on the subset of the first plurality of candidate parameters is less than the result of the cost function based on the subset of the second plurality of parameters, replacing the subset of the second plurality of parameters with the subset of the first plurality of candidate parameters in the HM; and in response to a determination that the result of the cost function based on the subset of the first plurality of candidate parameters is not less than the result of the cost function based on the subset of the second plurality of parameters, removing the subset of the first plurality of candidate parameters.

15. The heat exchanging system of claim 14, wherein the PAR is given by x.sub.new=x.sub.old+FW.Math.ϵ, wherein x.sub.old denotes a current pitch, FW is a fret width, ϵ∈(0,1] is a random number, and x.sub.new denotes an adjusted pitch.

16. The heat exchanging system of claim 15, wherein $FW\left(i\right)=F{W}_{\max }\exp \left(\ln \left(\frac{F{W}_{\min }}{F{W}_{\max }}\right)*\frac{i}{\mathrm{NI}}\right),$ wherein FW.sub.max denotes a maximum fret width, FW.sub.min denotes a minimum fret width, NI is a total number of the iterations, and i is a current iteration number.

17. A method of controlling a counter flow double pipe heat exchanger (DPHE) that includes a hot fluid pipe and a cold fluid pipe, the method comprising: determining a temperature error between a reference temperature and a temperature at an outlet of the hot fluid pipe of the counter flow DPHE; determining a change rate of the temperature error; determining a cold fluid mass flow rate from an output of a fuzzy logic controller based on the temperature error and the change rate of the temperature error being input to the fuzzy logic controller, the cold fluid mass flow rate being used for a cold fluid in the cold fluid pipe of the counter flow DPHE; and controlling the temperature within a predefined range by utilizing parameters of the fuzzy logic controller that are set based on a minimization of a cost function.

18. The method of claim 17, wherein the output of the fuzzy controller is limited by a function given $\Psi \left(u\right)=\{\begin{array}{cc}2,& u>2\\ u,& 0\le u\le 2\\ 0,& u<0\end{array},$ by wherein u denotes the output of the fuzzy controller.

19. A heat exchanging system, comprising: a counter flow double pipe heat exchanger (DPHE) configured to perform a heat exchanging process between a hot fluid in a hot fluid pipe and a cold fluid in a cold fluid pipe; and a fuzzy logic controller configured to: receive (i) a temperature error between a reference temperature and a temperature at an outlet of the hot fluid pipe of the counter flow DPHE and (ii) a change rate of the temperature error, and output a cold fluid mass flow rate for the cold fluid in the cold fluid pipe of the counter flow DPHE, wherein the temperature error is controlled within a predefined range by utilizing parameters of the fuzzy logic controller that are set based on a minimization of a cost function.

20. The heat exchanging system of claim 19, wherein the output of the fuzzy controller is limited by a function given by $\Psi \left(u\right)=\{\begin{array}{cc}2,& u>2\\ u,& 0\le u\le 2\\ 0,& u<0\end{array},$ wherein u denotes the output of the fuzzy controller.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0017] A more complete appreciation of this disclosure and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings, wherein:

[0018] FIG. 1 shows a simplified counter flow double pipe heat exchanger (DPHE) according to some embodiments of the disclosure;

[0019] FIG. 2 shows a cross section view of the counter flow DPHE according to some embodiments of the disclosure;

[0020] FIG. 3 shows an inner pipe fouling build-up pattern of the counter flow DPHE according to some embodiments of the disclosure;

[0021] FIG. 4 shows an open-loop response of the counter flow DPHE according to some embodiments of the disclosure;

[0022] FIG. 5 shows an exemplary block diagram of a proportional-integral-derivative (PID) controlled counter flow DPHE system according to some embodiments of the disclosure;

[0023] FIG. 6 shows a flowchart of a harmony search algorithm (HSA) according to some embodiments of the disclosure;

[0024] FIG. 7 shows an exemplary architecture of a fuzzy controller according to some embodiments of the disclosure;

[0025] FIG. 8 shows an exemplary architecture of a fuzzy PID (FPID) controlled counter flow DPHE system according to some embodiments of the disclosure;

[0026] FIGS. 9A-9B show the membership functions for the inputs of the fuzzy controller according to some embodiments of the disclosure;

[0027] FIG. 10 shows the membership functions for the output of the fuzzy controller according to some embodiments of the disclosure;

[0028] FIG. 11 shows the HSA convergence according to some embodiments of the disclosure;

[0029] FIG. 12 shows the hot fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under nominal conditions, respectively, according to some embodiments of the disclosure;

[0030] FIGS. 13-14 illustrate the cold fluid mass flow rates for the PID and FPID controlled counter flow DPHE systems under nominal conditions, respectively, according to some embodiments of the disclosure;

[0031] FIG. 15 shows the cold fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under nominal conditions, respectively, according to some embodiments of the disclosure;

[0032] FIG. 16 shows the hot fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under varying reference temperature conditions, respectively, according to some embodiments of the disclosure;

[0033] FIG. 17 shows the cold fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under varying reference temperature conditions, respectively, according to some embodiments of the disclosure;

[0034] FIGS. 18-19 show the cold fluid mass flow rates for the PID and FPID controlled counter flow DPHE systems under varying reference temperature conditions, respectively, according to some embodiments of the disclosure;

[0035] FIG. 20 shows a varied inner pipe hot fluid mass flow rate according to some embodiments of the disclosure;

[0036] FIG. 21 shows the hot fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under varying the hot fluid mass flow rate conditions, respectively, according to some embodiments of the disclosure;

[0037] FIGS. 22-23 show the cold fluid mass flow rates for the PID and FPID controlled counter flow DPHE systems under varying the hot fluid mass flow rate conditions, respectively, according to some embodiments of the disclosure;

[0038] FIG. 24 shows the cold fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under varying the hot fluid mass flow rate conditions, respectively, according to some embodiments of the disclosure;

[0039] FIG. 25 shows a varied hot fluid inlet temperature and a varied cold fluid inlet temperature, respectively, according to some embodiments of the disclosure;

[0040] FIG. 26 shows the hot fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under varying inlet temperature conditions, respectively, according to some embodiments of the disclosure;

[0041] FIGS. 27-28 show the cold fluid mass flow rates for the PID and FPID controlled counter flow DPHE systems under varying the inlet temperature conditions, respectively, according to some embodiments of the disclosure;

[0042] FIG. 29 shows the cold fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under varying the inlet temperature conditions, respectively, according to some embodiments of the disclosure;

[0043] FIG. 30 shows the hot fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under fouling build-up conditions, respectively, according to some embodiments of the disclosure;

[0044] FIGS. 31-32 show the cold fluid mass flow rates for the PID and FPID controlled counter flow DPHE systems under fouling build-up conditions, respectively, according to some embodiments of the disclosure;

[0045] FIG. 33 shows the cold fluid outlet temperatures for the PID and FPID controlled counter flow DPHE systems under fouling build-up conditions, respectively, according to some embodiments of the disclosure;

[0046] FIG. 34 shows a flowchart of operating the PID controlled counter flow DPHE system according to some embodiments of the disclosure;

[0047] FIG. 35 shows a flowchart of operating the FPID controlled counter flow DPHE system according to some embodiments of the disclosure;

[0048] FIG. 36 shows a simplified artificial neuron network (ANN) architecture according to some embodiments of the disclosure;

[0049] FIG. 37 shows an exemplary implemented ANN architecture according to some embodiments of the disclosure;

[0050] FIG. 38 shows an exemplary long short-term memory block according to some embodiments of the disclosure;

[0051] FIG. 39 shows an exemplary model to predict the tube fouling in a shell and tube heat exchanger (STHX) according to some embodiments of the disclosure;

[0052] FIG. 40 shows a steady state simulation of the STHX according to some embodiments of the disclosure;

[0053] FIG. 41 shows the ANN performance on a test set according to some embodiments of the disclosure; and

[0054] FIG. 42 shows a prediction performance of the fouling factor according to some embodiments of the disclosure.

DETAILED DESCRIPTION

[0055] In the drawings, like reference numerals designate identical or corresponding parts throughout the several views. Further, as used herein, the words “a,” “an” and the like generally carry a meaning of “one or more,” unless stated otherwise. The drawings are generally drawn to scale unless specified otherwise or illustrating schematic structures or flowcharts.

[0056] Furthermore, the terms “approximately,” “approximate,” “about,” and similar terms generally refer to ranges that include the identified value within a margin of 20%, 10%, or preferably 5%, and any values there between.

[0057] Aspects of this disclosure are directed to systems, devices, and methods for controlling a counter flow double pipe heat exchanger (DPHE). Two control schemes are included in the disclosure. The first one is a proportional-integral-derivative (PID) controller utilizing a harmony search metaheuristic algorithm developed to control an inner pipe fluid outlet temperature of a counter-flow heat exchanger undergoing fouling. The second control scheme is a fuzzy PID (FPID) controller to control the inner pipe fluid outlet temperature of the counter-flow heat exchanger undergoing fouling.

1. Modelling of the Counter Flow DPHE

[0058] FIG. 1 shows a simplified counter flow DPHE according to some embodiments of the disclosure. A one-dimensional finite-difference analysis can be applied to the counter flow DPHE. In the one-dimensional finite-difference analysis, the DPHE is partitioned into multiple partitions along the pipe length of the DPHE. In FIG. 1, T.sub.h.sub.in(t) is the hot side inlet temperature, T.sub.h.sub.out(t) is the hot side outlet temperature, T.sub.C.sub.in(t) is the cold side inlet temperature, T.sub.C.sub.out(t) is the cold side outlet temperature, L is a length of pipe, M is the number of the partitions, Δx is a length of each partition, and i is a spatial index.

[0059] FIG. 2 shows a cross section view of the counter flow DPHE according to some embodiments of the disclosure. In FIG. 2, d.sub.i is the inner diameter of the inner pipe, d.sub.o is the outer diameter of the inner pipe, D.sub.i is the inner diameter of the outer pipe, and D.sub.o is the outer diameter of the outer pipe.

[0060] Based on the one-dimensional finite-difference analysis of the counter flow DPHE, a dynamic DPHE model can be derived. The following assumptions can be used for the model derivations:

[0061] (1) The heat transfer coefficient is uniform along the pipe length;

[0062] (2) The wall temperature is considered negligible;

[0063] (3) The density and heat capacity of hot and cold fluids are constants; and

[0064] (4) The hot and cold fluids are both in a liquid phase.

[0065] As described by F. P. Incropera et al. in “Fundamentals of heat and mass transfer,” John Wiley & Sons, 2011, the first law of the thermodynamic (conservation of energy) can be expressed as,

[00001]$\begin{array}{cc}{\stackrel{.}{E}}_{\mathrm{in}}-{\stackrel{.}{E}}_{\mathrm{out}}={\stackrel{.}{E}}_{\mathrm{st}}=\rho C_{P}V\frac{\partial T}{\partial t}& \left(1\right)\end{array}$

where Ė.sub.in and Ė.sub.out denote the energy transfer rates into and out of the heat exchanger, respectively. The rate of the energy change of the system is represented by Ė.sub.st.

[0066] In some embodiments, it is assumed that the hot fluid flows through the inner pipe and the cold fluid flows through the outer pipe of the heat exchanger, equation (1) can be rewritten as follows:

[00002]$\begin{array}{cc}{\rho }_h{C}_{P_h}\Delta V_1\frac{\partial T_h\left(t,i\right)}{\partial t}={\stackrel{.}{m}}_h{C}_{P_h}\left(T_h\left(t,i-1\right)-T_h\left(t,i\right)\right)-U\left(t\right)\Delta A_s\left(i\right)\left(T_h\left(t,i\right)-T_c\left(t,i\right)\right)& \left(2\right)\\ {\rho }_c{C}_{P_c}\Delta V_2\frac{\partial T_c\left(t,i\right)}{\partial t}={\stackrel{.}{m}}_c{C}_{P_c}\left(T_c\left(t,i-1\right)-T_c\left(t,i\right)\right)+U\left(t\right)\Delta A_s\left(i\right)\left(T_h\left(t,i\right)-T_c\left(t,i\right)\right)& \left(3\right)\end{array}$

where {dot over (m)}.sub.h is the hot fluid mass flow rate (Kg/sec), {dot over (m)}.sub.c is the cold fluid mass flow rate (Kg/sec), ρ.sub.h is the hot fluid density (Kg/m.sup.3), ρ.sub.c is the cold fluid density (Kg/m.sup.3), U(t) is the overall heat coefficient (W/m.sup.2.Math.K), ΔA.sub.s is the change in the surface area (m.sup.2), ΔV.sub.1 is the change in the inner pipe volume (m.sup.3), ΔV.sub.2 is the change in the outer pipe volume (m.sup.3), Δt is the time step (sec), C.sub.P.sub.h is the specific heat at a constant pressure of the hot fluid, and C.sub.P.sub.c is the specific heat at a constant pressure of the cold fluid.

[0067] By using explicit discretization and using an upwind scheme for the convective terms, the following equations can be obtained:

[00003]$\begin{array}{cc}\frac{T_h\left(i,t+\Delta t\right)-T_h\left(i,t\right)}{\Delta t}=\frac{{\stackrel{.}{m}}_h{C}_{P_h}\left(T_h\left(t,i-1\right)-T_h\left(t,i\right)\right)-U\left(t\right)\Delta A_s\left(i\right)\left(T_h\left(t,i\right)-T_c\left(t,i\right)\right)}{{\rho }_h{C}_{P_h}\Delta V_1}& \left(4\right)\\ \frac{T_c\left(i,t+\Delta t\right)-T_c\left(i,t\right)}{\Delta t}=\frac{{\stackrel{.}{m}}_c{C}_{P_c}\left(T_c\left(t,i-1\right)-T_c\left(t,i\right)\right)+U\left(t\right)\Delta A_s\left(i\right)\left(T_h\left(t,i\right)-T_c\left(t,i\right)\right)}{{\rho }_c{C}_{P_c}\Delta V_2}& \left(5\right)\end{array}$

[0068] After some arrangements, Equations (4) and (5) can be written as follows:

[00004]$\begin{array}{cc}{T}_h\left(i,t+\Delta t\right)=T_h\left(i,t\right)+\frac{\begin{array}{c}\Delta t({\stackrel{.}{m}}_h{C}_{P_h}\left(T_h\left(t,i-1\right)-T_h\left(t,i\right)\right)-\\ U\left(t\right)\Delta A_s\left(i\right)\left(T_h\left(t,i\right)-T_c\left(t,i\right)\right))\end{array}}{{\rho }_c{C}_{P_c}\Delta V_1}& \left(6\right)\\ T_c\left(i,t+\Delta t\right)=T_c\left(i,t\right)+\frac{\begin{array}{c}\Delta t({\stackrel{.}{m}}_c{C}_{P_c}\left(T_c\left(t,i-1\right)-T_c\left(t,i\right)\right)+\\ U\left(t\right)\Delta A_s\left(i\right)\left(T_h\left(t,i\right)-T_c\left(t,i\right)\right))\end{array}}{{\rho }_c{C}_{P_c}\Delta V_2}& \left(7\right)\\ \mathrm{where}\Delta A_s=\pi d_o\Delta x,\Delta V_1=\frac{\pi d_o^2}{4}\Delta x,\mathrm{and}\Delta V_2=\frac{\pi D_i^2}{4}\Delta x& \left(8\right)\end{array}$

[0069] In some embodiments, the fouling has been considered for the inner pipe only while the outer pipe fouling is neglected. The thermal effect of the fouling on the heat transfer coefficient can be represented as follows:

[00005]$\begin{array}{cc}{U}_f\left(t\right)=\frac{U_c\left(t\right)}{1+U_c{R}_f}& \left(9\right)\end{array}$

where U.sub.ƒ is the heat transfer coefficient under fouling conditions, U.sub.c is the heat transfer coefficient in the clean state, and R.sub.ƒ is the fouling resistance.

[0070] FIG. 3 shows an inner pipe fouling build-up pattern according to some embodiments of the disclosure. Given the fouling regime shown in FIG. 3, a resultant open-loop response is illustrated in FIG. 4 according to some embodiments of the disclosure.

[0071] Table 1 shows the DPHE parameters used for the system simulation according to some embodiments of the disclosure.

TABLE-US-00001 TABLE 1 Hot fluid mass flow rate ({dot over (m)}.sub.h) 0.1 Kg/sec Cold fluid mass flow rate ({dot over (m)}.sub.c) 0.0552 Kg/sec Specific heat at a constant pressure of 4180 J/Kg .Math. K hot fluid (C.sub.ph) Specific heat at a constant pressure of 4180 J/Kg .Math. K cold fluid (C.sub.pc) Hot fluid density (ρ.sub.h) 1000 Kg/m.sup.3 Cold fluid density (ρ.sub.c) 1000 Kg/m.sup.3 Length of pipe 5 m Overall heat coefficient (U) 1000 W/n.sup.2 .Math. K Number of segments (M) 50 Time step 0.01 sec Hot side inlet temperature 100° C. Cold side inlet temperature  20° C. Inner diameter of the inner pipe (d.sub.i) 0.04 m Outer diameter of the inner pipe (d.sub.o) 0.05 m Inner diameter of the outer pipe (D.sub.i) 0.1 m

2. Control Design

[0072] In this disclosure, two feedback control schemes have been employed to perform the fouling compensation based on the presented numerical model of the counter flow DPHE. In a first control scheme, a PID controller utilizing a harmony search metaheuristic algorithm is used to control the cold fluid flow rate of the counter flow DPHE based on temperature error between a reference temperature and a hot side outlet temperature (e.g., T.sub.h.sub.out(t) in FIG. 1) of the counter flow DPHE. In a second control scheme, a fuzzy PID (FPID) controller is adopted to control the cold fluid flow rate of the counter DPHE based on the temperature error between the reference temperature and the hot side outlet temperature (e.g., T.sub.h.sub.out(t) in FIG. 1) of the counter flow DPHE and a change rate of the temperature error.

2.1. PID Parameters Using Harmony Search Algorithm

[0073] The PID controller is known for its simplicity, ease of realization, and functionality, so it has been extensively used in various industrial systems, as described by K. H. Ang et al. in “PID control system analysis, design, and technology,” IEEE Trans. Control Syst. Technol., vol. 13, no. 4, pp. 559-576, 2005. Several Model-based and model-free tuning methods have been developed to tune the PID controller, as described by K. Ogata et al. in “Modern control engineering,” 5th ed., vol. 5, Prentice Hall Upper Saddle River, 2010. However, these tuning strategies might not lead to an optimal performance, especially when the controlled plant is nonlinear.

[0074] Based on the above presented discrete-time model of the DPHE, a discrete PID controller has been utilized to manipulate the cold fluid mass flow rate according to the desired hot fluid outlet temperature (e.g., the reference temperature). A mathematical formula for the discrete-time PID controller is given by:

[00006]$\begin{array}{cc}{u}_t=\left(k_p+k_i\frac{T_s}{1-z^{-1}}+k_d\frac{1}{T_s}\left(1-z^{-1}\right)\right)e_t& \left(10\right)\end{array}$

where u.sub.t is the output of the PID controller, k.sub.p is the proportional gain, k.sub.d is the derivative gain, k.sub.i is the integral gain, e denotes the error, T.sub.s represents the sampling time, and z.sup.−1 is a backward shift operator, as described by E. M. Shaban et al. in “A novel discrete PID+ controller applied to higher order/time delayed nonlinear systems with practical implementation,” Int. J. Dyn. Control, vol. 7, no. 3, pp. 888-900, 2019.

[0075] FIG. 5 illustrates a block diagram of the PID-controlled counter flow DPHE according to some embodiments of the disclosure.

[0076] As described by Z. W. Geem et al. in “A new heuristic optimization algorithm: Harmony search,” Simulation, vol. 76, no. 2, pp. 60-68, 2001, the harmony search algorithm (HSA) is a musically inspired meta-heuristic algorithm. The HSA is similar to music improvisation. A harmony memory (HM) is a location where the HSA saves a solution set. A new harmony in the HSA can be chosen from the HM or from the HM with slight adjustments, or it can be selected at random within the possible solution zone, as described by Z. W. Geem in “Music-inspired harmony search algorithm,” which was published on vol. 191, no. 1. 2009, and by T. Zhang et al. in “Review of harmony search with respect to algorithm structure,” Swarm Evol. Comput., vol. 48, pp. 31-43, 2019. Pitch adjusting rate (PAR) and harmony memory considering rate (HMCR) are two parameters that guide the selection process. PAR∈[0,1], which represents a slight adjustment to the past value in the HM, is given by:

x.sub.new=x.sub.old+FW.Math.ϵ  (11)

where x.sub.new represents the adjusted pitch, x.sub.old denotes the current solution or pitch, FW is a fret width, and ϵ∈[0,1] is a random number.

[0077] HMCR ∈[0,1] represents a likelihood of selecting a proposed solution from among the current members of the HM. Picking up tiny HMCR values for the harmony can result in a slow convergence for the selection process. On the other hand, picking up high HMCR values (around 1) can result in a faster convergence but may compromise algorithm exploration capabilities.

[0078] FIG. 6 shows a flowchart of the HSA according to some embodiments of the disclosure. As described by T. Zhang et al. in “Review of harmony search with respect to algorithm structure,” Swarm Evol. Comput., vol. 48, pp. 31-43, 2019, and by K. S. Lee et al. in “A new meta-heuristic algorithm for continuous engineering optimization: Harmony search theory and practice,” Comput. Methods Appl. Mech. Eng., vol. 194, no. 36-38, pp. 3902-3933, 2005, the procedure of the HSA includes the following steps:

[0079] Step S610: Harmony memory initialization, where initial candidate solutions in the HM are produced randomly. An HM with a size of HMS can be implemented as:

[00007]$\begin{array}{cc}\mathrm{HM}=\left[\begin{array}{cccc}{x}_1^1& x_2^1& & x_n^1\\ x_1^2& x_2^2& \mathrm{.Math.}& x_n^2\\ & \mathrm{.Math.}& \ddots & \mathrm{.Math.}\\ x_1^{\mathrm{HMS}}& x_2^{\mathrm{HMS}}& \mathrm{.Math.}& x_n^{\mathrm{HMS}}\end{array}\right]& \left(12\right)\end{array}$

where [x.sub.1.sup.i, x.sub.2.sup.i, . . . , x.sub.n.sup.i] are the candidate solutions (i=1, 2, . . . , HMS).

[0080] Step S620: Improvise a new harmony from the HM in accordance with the HMCR and PAR.

[0081] Step S630: Evaluate the new harmony.

[0082] Step S640: Compare the new harmony value with the worst harmony value stored in the HM. If the new harmony value fits better, proceed to step S650; otherwise, the new harmony will be removed.

[0083] Step S650: Update the HM by replacing the worst harmony with the new harmony.

[0084] Step S660: Determine whether the termination criteria are satisfied. If not, return to step S620.

[0085] The implemented algorithm is inspired by the original HSA approach. However, instead of producing only one harmony per improvisation, multiple harmonics are created to improve the convergence rate, as described by Y. Cheng et al. in “An improved harmony search minimization algorithm using different slip surface generation methods for slope stability analysis,” Eng. Optim., vol. 40, no. 2, pp. 95-115, 2008. In addition, as described by M. Mahdavi et al. in “An improved harmony search algorithm for solving optimization problems,” Appl. Math. Comput., vol. 188, no. 2, pp. 1567-1579, 2007, a variable FW has been applied as follows,

[00008]$\begin{array}{cc}FW\left(i\right)=F{W}_{\max }\exp \left(\ln \left(\frac{F{W}_{\min }}{F{W}_{\max }}\right)*\frac{i}{\mathrm{NI}}\right)& \left(13\right)\end{array}$

where FW.sub.max indicates the maximum fret width, FW.sub.min indicates the minimum fret width, NI is the total generations number, and i is the current generation number. Initially, FW has a high value for improving global search capacity. To increase local search capabilities, FW steadily declines as i increases.

2.2. Fuzzy-PID Controller

[0086] The fuzzy logic control is one of the successful applications of fuzzy set theory, which, unlike crisp sets, allows partial membership (i.e., an element can be in part a member of more than a single set at the same time). The fuzzy logic receives an input vector and maps it into an output vector using linguistic statements in a form of If-Then rules attained from utilizing human knowledge about the system, as described by K. M. Passino and S. Yurkovich in “Fuzzy control”, Vol. 42, Addison Wesley, 1998.

[0087] FIG. 7 shows an exemplary architecture of a fuzzy controller according to some embodiments of the disclosure. The operation of the fuzzy control system can include four stages as follows:

[0088] (1) Fuzzification, wherein the controller inputs are mapped into linguistic expressions.

[0089] (2) Rule-base, which consists of rules in IF-THEN format.

[0090] (3) Inference engine infers fuzzy control action from the knowledge of linguistic variables and control rules.

[0091] (4) Defuzzification, wherein the fuzzy linguistic output is transferred into a numerical output.

[0092] As described by J. M. Mendel in “Uncertain rule-based fuzzy systems,” Springer International Publishing 2017, a fuzzy PID (FPID) controller has been designed to control the DPHE inner pipe outlet temperature, which represents the hot fluid outlet temperature in some embodiments. The error between the desired temperature and the actual hot fluid outlet temperature, as well as the rate of change of this error, represent the inputs of the controller, while the output of the controller is the outer pipe fluid (cold fluid) mass flow rate {dot over (m)}.sub.c.

[0093] FIG. 8 shows an exemplary FPID controlled counter flow DPHE system according to some embodiments of the disclosure. The FPID controlled counter flow DPHE system is based on Takagi-Sugeno fuzzy systems, where Ψ(u), which limits the controller output, is expressed as follows:

[00009]$\begin{array}{cc}\Psi \left(u\right)=\{\begin{array}{cc}2,& u>2\\ u,& 0\le u\le 2\\ 0,& u<0\end{array}& \left(14\right)\end{array}$

[0094] In FIG. 8, the parameters K.sub.e and K.sub.d are positive constants that control the controller sensitivity to the error (e) and the change in the error (de), respectively. K.sub.o is an output scaling factor, and K.sub.I is an integral gain.

[0095] Triangular membership functions are adopted for fuzzy inputs and output with an input universe of discourse as [Negative (N), Zero (Z), Positive (P)]. The output universe of discourse is [Negative Big (NB), Negative Medium (NM), Zero (Z), Positive Medium (PM), Positive Big (PB)].

[0096] FIGS. 9A-9B show the membership functions for the inputs (the error and the change in the error) of the fuzzy controller according to an embodiment of the disclosure.

[0097] FIG. 10 shows the membership functions for the output of the fuzzy controller according to an embodiment of the disclosure.

[0098] Table 2 illustrates the rules used to design the fuzzy controller according to some embodiments of the disclosure.

TABLE-US-00002 TABLE 2 ė {dot over (m)}.sub.c N Z P e N NB NM Z Z NM Z P P Z PM PB

3. Simulation Results

[0099] According to aspects of the disclosure, the HSA is used to tune the PID controller parameters K.sub.p,i, and K.sub.d, by minimizing a cost function as expressed in equation (15), which is defined based on the Integral of Time multiplied Absolute Error (ITAE) and the standard deviation of the output vector of the controller, where the latter index is used to penalize the chattering in the controller output.

[00010]$\begin{array}{cc}\mathrm{cost}\mathrm{function}=\frac{1}{N}{.Math.}_{k=1}^{N}k.Math.ϵ\left(k\right).Math.+{10}^3*\sqrt{\frac{1}{N-1}{.Math.}_{k=1}^N{\left(u\left(k\right)-u\right)}^2}& \left(15\right)\end{array}$

[0100] where 0≤u(k)≤2, 0≤K.sub.p≤5, 0≤K.sub.d≤10, 0≤K.sub.i≤10, k represents a time index and N denotes the number of samples while E is the error between actual hot fluid outlet temperature and the desired temperature (ϵ=T.sub.h.sub.out−T.sub.d) while u denotes the mean value of the output vector of the fuzzy controller. The error at the steady state is more penalized than the error at the beginning because of the time multiplication term. The DPHE system is considered at its standard conditions, as shown in Table 1, where the desired temperature is 60° C., and the simulation time is set to 3 hours for example.

[0101] FIG. 11 shows the HSA convergence according to some embodiments of the disclosure.

[0102] Table 3 shows the optimized parameters of the PID controller according to some embodiments of the disclosure.

TABLE-US-00003 TABLE 3 Parameter K.sub.p K.sub.i K.sub.d value 0.6553 0.0107 9.99

[0103] Table 4 shows the parameters of the fuzzy controller that is implemented according to the structure in FIG. 8.

TABLE-US-00004 TABLE 4 Parameter K.sub.p K.sub.i K.sub.d K.sub.o value 0.6553 0.0321 9.99 2

[0104] Five scenarios have been simulated for evaluating both the PID and fuzzy controllers to show the robustness of the designed controllers under different operation conditions where the controlled DPHE system has been simulated using SIMULINK.

3.1. Scenario 1: The Operation under Nominal Conditions

[0105] In this scenario, the simulation is conducted based on the DPHE nominal parameters given in Table 1, where the reference outlet temperature of the hot fluid (inner pipe fluid) has been adjusted to 60° C. FIG. 12 shows the hot fluid outlet temperatures for both controllers, where the FPID controller exhibits a better dynamic performance with a faster settling time. The PID and FPID controllers recorded an ITAE of 100 and 83.6287, respectively. FIGS. 13-14 illustrate the cold fluid mass flow rate control signals of the PID and FPID controllers, respectively, where both FPID and PID controllers show smooth outputs. The cold fluid outlet temperatures of the PID and FPID are depicted in FIG. 15.

3.2. Scenario 2: Varying Reference Temperature for Inner Pipe Outlet Temperature

[0106] This scenario evaluates the dynamics of the controlled DPHE system subjected to a varying reference temperature. The simulation is conducted for five hours, where the reference temperature is altered from 60° C. to 55° C., then to 70° C., and finally returned to 60° C. at the end of simulation time. FIG. 16 depicts the tracking performances of the PID and FPID controllers, whereas both controllers experience abrupt changes during the transition period from one reference to another. FIG. 17 shows the cold fluid outlet temperatures of the PID and FPID controllers under varying reference temperature conditions. The FPID controller proves a better response with ITAE equals to 1351, while the PID controller shows an ITAE of 1684. Both the FPID and PID controller outputs are shown in FIGS. 18-19, respectively.

3.3. Scenario 3: Varying Inner Pipe Inlet Flow Rate

[0107] The inner pipe (hot) fluid mass flow rate has been varied, as illustrated in FIG. 20. Both controllers have absorbed the effect of change in mass flow rate, as demonstrated in FIG. 21. The PID controller output is depicted in FIG. 22, while the outer pipe mass flow rate obtained by the FPID controller is shown in FIG. 23. The FPID and PID controllers recorded an ITAE of 85.8653 and 112.2291, respectively, which reveals that the FPID performance is better than PID in this scenario. The cold fluid outlet temperatures for the PID and FPID controllers under varying the hot fluid mass flow rate conditions are demonstrated in FIG. 24.

3.4. Scenario 4: Varying Inlet Temperatures

[0108] This scenario investigates the effect of change in the inlet temperatures of the feedback controlled DPHE on the inner outlet temperature regulation. The hot fluid inlet temperature has been altered within the range 95° C. to 105° C.; at the same time, the cold fluid inlet temperature has been varied between 17° C. and 23° C., as illustrated in FIG. 25.

[0109] FIG. 26 shows the inner pipe outlet temperatures for the PID and FPID controllers, where both controllers offer satisfactory responses. ITAE for the FPID controller is 31.4135, while the PID controller has a higher ITAE of 36.16. FIGS. 27-28 depict the mass flow rate control signals obtained from the PID controller and FPID, respectively. The outer pipe outlet temperatures are illustrated in FIG. 29.

3.5. Scenario 5: Fouling Build-up

[0110] In this scenario, the fouling build-up has been deemed where the fouling resistance is varied, as demonstrated in FIG. 3, to investigate the fouling influence on the FPID and PID controller performances. The build-up of the fouling starts after 5 hours and continues until it reaches the peak value at the end of the simulation period, which is equivalent to a reduction in the heat transfer coefficient by 50%. Both controllers demonstrated excellent regulation capability, as shown in FIG. 30. The FPID controller shows an ITAE value equals 5.17 versus 5.4 for the PID controller.

[0111] Table 5 compares the PID and FPID controlled DPHE errors based on IATE and integral square error (ISE) indexes. Based on the comparison, it is apparent that the FPID controlled DPHE has a better performance for all considered operating scenarios.

TABLE-US-00005 TABLE 5 IATE IATE ISE ISE (PID) (FPID) (PID) (PID) Scenario 1 100 83.63 10.90 10.71 Scenario 2 1684.9 1351.2 4.26 3.81 Scenario 3 112.22 85.87 2.46 2.39 Scenario 4 36.16 34.41 1.74 1.70 Scenario 5 5.41 5.17 0.20 0.20

[0112] FIGS. 31-32 show the cold fluid mass flow rates for both PID controller and FPID, correspondingly, while FIG. 33 depicts the cold fluid outlet temperatures of the DPHE using the PID and FPID controllers under fouling build-up conditions, respectively.

4. Flowcharts of Operating the DPHE

[0113] FIG. 34 shows a flowchart of a process to operate the PID controlled counter flow DPHE system according to some embodiments of the disclosure.

[0114] In step S3410, the process determines a temperature error between a reference temperature and a temperature at an outlet of the hot fluid pipe of the counter flow DPHE.

[0115] In step S3420, the process determines a cold fluid mass flow rate from an output of a PID controller based on the temperature error being input to the PID controller, the cold fluid mass flow rate being used for a cold fluid in the cold fluid pipe of the counter flow DPHE.

[0116] In step S3430, the process controls the temperature error within a predefined range by utilizing parameters of the PID controller that are set by using a harmony search algorithm (HSA) to obtain a minimization of a cost function.

[0117] In an embodiment, the hot fluid pipe and the cold fluid pipe are an inner pipe and an outer pipe of the counter flow DPHE, respectively.

[0118] In an embodiment, the parameters of the PID controller include a proportional gain K.sub.p, a derivative gain K.sub.d, and an integral gain K.sub.i.

[0119] In an embodiment, 0≤K.sub.p≤5, 0≤K.sub.d≤10, and 0≤K.sub.i≤10.

[0120] In an embodiment, the cost function is given by

[00011]$\mathrm{cost}\mathrm{function}=\frac{1}{N}{.Math.}_{k=1}^{N}k.Math.ϵ\left(k\right).Math.+{10}^3*\sqrt{\frac{1}{N-1}{.Math.}_{k=1}^N{\left(u\left(k\right)-u\right)}^2},$

wherein N denotes a total number of time samples during the minimization of the cost function, k denotes a time index, ϵ denotes the temperature error, u∈[0,2] denotes the output of the PID controller, and μ denotes a mean value of the outputs of the PID controller during the minimization of the cost function.

[0121] In an embodiment, the minimization of the cost function includes multiple iterations, wherein each iteration includes: (i) generating a first plurality of candidate parameters from a second plurality of parameters according to a pitch adjusting rate (PAR) and a harmony memory considering rate (HMCR) of the HSA, the second plurality of parameters being stored in a harmony memory (HM); (ii) determining whether a result of the cost function based on a subset of the first plurality of candidate parameters is less than a result of the cost function based on a subset of the second plurality of parameters; (iii) in response to a determination that the result of the cost function based on the subset of the first plurality of candidate parameters is less than the result of the cost function based on the subset of the second plurality of parameters, replacing the subset of the second plurality of parameters with the subset of the first plurality of candidate parameters in the HM; and (iv) in response to a determination that the result of the cost function based on the subset of the first plurality of candidate parameters is not less than the result of the cost function based on the subset of the second plurality of parameters, removing the subset of the first plurality of candidate parameters.

[0122] In an embodiment, the PAR is given by x.sub.new=x.sub.old+FW.Math.ϵ, wherein x.sub.old denotes a current pitch, FW is a fret width, ϵ∈(0,1] is a random number, and x.sub.new denotes an adjusted pitch.

[0123] In an embodiment,

[00012]$FW\left(i\right)=F{W}_{\max }\exp \left(\ln \left(\frac{F{W}_{\min }}{F{W}_{\max }}\right)*\frac{i}{\mathrm{NI}}\right),$

wherein FW.sub.max denotes a maximum fret width, FW.sub.min denotes a minimum fret width, NI is a total number of iterations, and i is a current iteration number.

[0124] FIG. 35 shows a flowchart of a process to operate the FPID controlled counter flow DPHE system according to some embodiments of the disclosure.

[0125] In step S3510, the process determines a temperature error between a reference temperature and a temperature at an outlet of the hot fluid pipe of the counter flow DPHE.

[0126] In step S3520, the process determines a change rate of the temperature error.

[0127] In step S3530, the process determines a cold fluid mass flow rate from an output of a fuzzy logic controller based on the temperature error and the change rate of the temperature error being input to the fuzzy logic controller, the cold fluid mass flow rate being used for a cold fluid in the cold fluid pipe of the counter flow DPHE.

[0128] In step S3540, the process controls the temperature within a predefined range by utilizing parameters of the fuzzy logic controller that are set based on a minimization of a cost function.

[0129] In an embodiment, the output of the fuzzy controller is limited by a function given by

[00013]$\Psi \left(u\right)=\{\begin{array}{cc}2,& u>2\\ u,& 0\le u\le 2\\ 0,& u<0\end{array},$

wherein u denotes the output of the fuzzy controller.

5. Heat Exchanger Fouling Prediction Using Artificial Intelligence

[0130] This disclosure also includes a data driven approach to predict the tube fouling in a shell and tube heat exchanger (STHX) so that the cleaning process of the fouling can be scheduled before the fouling reaches a severe level.

[0131] According to aspects of the disclosure, a feedforward artificial neural network (ANN) is used for real-time fouling estimation and a long short-term memory (LSTM) neural network is adopted for prediction purposes.

5.1. Mathematical Model

[0132] The STHX is one of the prevalent and common types of heat exchangers that are used extensively in a board range of industrial processes. It includes a tube bundle that contains one of a hot fluid and a cold fluid while the shell contains the other fluid.

[0133] As described by T. Ardsomang et al. in “Heat exchanger fouling and estimation of remaining useful life,” Proc. Annu. Conf. Progn. Heal. Manag. Soc., pp. 150-158, 2013, the heat transfer rate of the STHX can be written as follows,

Q=U.sub.oAΔT.sub.LMTDF   (16)

where A is the area of the heat transfer, U.sub.o is the overall coefficient of heat transfer, ΔT.sub.LMTD is the logarithmic mean temperature difference (LMTD) which is given by Equation (17), and F is the correction factor of LMTD.

[00014]$\begin{array}{cc}\Delta T_{LMTD}=\frac{\Delta T_1-\Delta T_2}{\ln \left(\Delta T_1\right)-\ln \left(\Delta T_2\right)}& \left(17\right)\end{array}$

[0134] In case of counter current flow

ΔT.sub.1=T.sub.hi−T.sub.co ΔT.sub.2=T.sub.ho−T.sub.ci   (18)

where h.sub.i and c.sub.i denote the inlet temperatures of the hot and cold fluids, respectively, while h.sub.o and c.sub.o denote the outlet temperatures of the hot and cold fluids, respectively.

[0135] The rate of the heat transfer can also be written based on energy balance as

Q=m.sub.hC.sub.P.sub.hΔT.sub.h=m.sub.cC.sub.P.sub.cΔT.sub.c   (19)

where C.sub.P.sub.h is the hot fluid specific heat capacity in J/(kg.Math.K), C.sub.P.sub.c is the cold fluid specific heat capacity in J/(kg.Math.K), m.sub.h is the hot fluid mass flow rate in kg/s, m.sub.c is the cold fluid mass flow rate in kg/s, ΔT.sub.h is the hot fluid temperature difference in K, and ΔT.sub.c is the cold fluid temperature difference in K.

[0136] The overall heat transfer coefficient is given by

[00015]$\begin{array}{cc}\frac{1}{U_o}=\frac{1}{U_c}+R_f& \left(20\right)\end{array}$

where U.sub.c is the heat transfer coefficient at clean state in

[00016]$\frac{W}{m^{2}K},$

and R.sub.ƒ is the fouling factor or resistance in

[00017]$\frac{m^{2}K}{W}.$

5.2. Artificial Neural Networks (ANN)

[0137] The ANN is a computational model inspirited by the biological neurons in the human nervous system. The ANN is one variant of data-driven models and it is known by its capability to extrapolate complex nonlinear relationships between inputs sequence and the corresponding output(s).

[0138] FIG. 36 shows a simplified ANN architecture according to some embodiments of the disclosure.

[0139] FIG. 37 shows an exemplary implemented ANN architecture according to some embodiments of the disclosure.

[0140] The implemented ANN architecture involves three layers: input layer, hidden layer, and output layer. Each layer contains several neurons, as described by S. Haykin in “Neural networks and learning machines, vol. 3” which was published by Pearson Prentice Hall New Jersey in 2008. Input and output layer dimensions are determined according to the given input and output whereas the hidden layer dimension depends on the complexity of the required task. Each neuron performs a weighted sum operation on its inputs and passes this summation through a differentiable activation function which can be either a linear or a nonlinear function. The output of the neuron is given by

h=ƒ(W.sup.TX+b)   (21)

where h is the output of the neuron, X is the input vector of the neuron, W is a weight vector, b is a constant representing a bias of the neuron, and ƒ is an activation function, as described by M. Gopal in “Applied Machine Learning,” McGraw-Hill Education in 2018.

[0141] The ANN complexity is proportional to the number of neurons in hidden layer besides the number of hidden layers. If there is more than one hidden layer in the ANN architecture, then it is referred to as a deep neural network.

[0142] Typically, given the input and the target output as in classification or regression problems, the ANN training process is performed using gradient decent or one of its variants as an optimization algorithm while the weight update is accomplished via a backpropagation algorithm based on a pre-defined objective function. The ANN has several hyper-parameters that affect the training performance such as the number of hidden layers, the number of hidden neurons, the learning rate, and the activation function, as described by M. Z. Alom et al. in “A state-of-the-art survey on deep learning theory and architectures,” Electron, vol. 8, no. 3, 2019.

5.3. Long Short-Term Memory (LSTM)

[0143] Conventional neural networks are independent of previous state values so that they cannot be used for prediction purposes. Recurrent neural networks (RNN) were developed to handle dynamic systems by adding a feedback path such that the neural networks output become dependent on the previous states. However, gradient vanishing problem in addition to difficulty in processing long sequences limits the RNN performance especially in time-series forecasting problems, as described by M. Abdel-Nasser et al. in “Accurate photovoltaic power forecasting models using deep LSTM-RNN,” Neural Comput. Appl., pp. 1-14, 2017.

[0144] Long short-term memory (LSTM) is developed to solve previously discussed RNN shortcomings with the ability to learn long and short-term dependencies. The LSTM encompasses a series connected memory blocks where each block as depicted in FIG. 38 consists of three gates: a forget gate (ƒ), an input gate (i), and an output gate (o).

[0145] Based on the pervious values of the hidden states h.sub.t−1 and X.sub.t. The forget gate evaluates which information of the cell state C.sub.t−1 to be thrown away. As described by C. Wang et al. in “Long short-term memory neural network (LSTM-NN) enabled accurate optical signal-to-noise ratio (OSNR) monitoring,” J. Light. Technol., vol. 37, no. 16, pp. 4140-4146, 2019, the block equation can be written as follows,

ƒt=σ(W.sub.ƒ[h.sub.t−1,x.sub.t]+b.sub.ƒ)   (22)

i.sub.t=σ(W.sub.i[h.sub.t−1,x.sub.t]+b.sub.i)   (23)

o.sub.t=σ(W.sub.o[h.sub.t−1,x.sub.t]+b.sub.o)   (24)

C.sub.t=ƒ.sub.t*C.sub.t−1+i.sub.t*tanh(W.sub.c[h.sub.t−1,x.sub.t]+b.sub.i)   (25)

h.sub.t=o.sub.t*tanh(C.sub.t)   (26)

where W.sub.i , W.sub.ƒ, W.sub.c, and W.sub.o represent the weight matrices while b.sub.i, b.sub.ƒ, b.sub.c, and b.sub.o denote biases and σ is a sigmoid activation function.

[0146] FIG. 39 shows an exemplary model consisting of two blocks to predict the tube fouling in the STHX according to some embodiments of the disclosure. The first block has a feedforward neural network to estimate the fouling while the second block is used to predict the fouling for several time steps ahead.

5.4. Simulation Results

[0147] FIG. 40 shows a steady state simulation of the STHX according to some embodiments of the disclosure. The steady state simulation is conducted in the PetroSIM software for an asymptotic fouling regime to get the training data for the fouling estimation and prediction. Water has been used for both the shell and the tube in the simulated STHX.

[0148] Table 6 shows the fluids and heat exchanger parameters according to some embodiments of the disclosure.

TABLE-US-00006 TABLE 6 Parameter Value Shell inner diameter 1.3 m Tuber inner diameter 21 mm Tube outer diameter 25 mm Number of Shell Passes 2 Number of tubes 1122 Tube Length 6.1 m Shell fluid inlet temperature 90° C. Tube fluid inlet temperature 30° C. Shell fluid mass flow rate 6.47 * 10.sup.8 Kg/h Tube fluid mass flow rate 1.368 * 10.sup.6 Kg/h

[0149] MATLAB deep learning toolbox has been adopted to implement the ANN with 20 hidden neurons as depicted in FIG. 37. The tube pressure drop as well as the shell and tube outlet temperatures can be selected as the ANN inputs while the output of the ANN is the fouling factor.

[0150] Prior to training the fouling estimation model, the obtained data was firstly normalized within a range (−1, 1) and then divided into three groups: training, validation, and testing. FIG. 41 shows the ANN performance on a test set according to some embodiments of the disclosure. The test set represents 15% of the total data where a very small error is obtained with MSE=8.5*10.sup.−11 where MSE is given by Equation 27.

[00018]$\begin{array}{cc}MSE=\frac{1}{N}{.Math.}_{i=1}^N{\left({\hat{Y}}_i-Y_i\right)}^2& \left(27\right)\end{array}$

where N is the number of data points, Ŷ is the estimated value, and Y is the observed value.

[0151] The estimated fouling factor obtained from the ANN represents the current input to the LSTM which has 20 hidden neurons. The LSTM is trained on the first 93% of the observed data that is obtained from the ANN estimation model. Then, the LSTM is used to predict the remaining 7%. FIG. 42 depicts the prediction performance where the LSTM showed a prediction error RMSE=1.72*10.sup.−6 which represents a small error considering the short term prediction. This performance allows this model to be used for heat exchanger predictive maintenance through providing early alarms about the status of the STHX.

[0152] All publications and patent applications mentioned in this specification are herein incorporated by reference in their entirety to the same extent as if each individual publication or patent application was specifically and individually indicated to be incorporated by reference, especially referenced is disclosure appearing in the same sentence, paragraph, page or section of the specification in which the incorporation by reference appears.