### Transcription

View metadata, citation and similar papers at core.ac.ukbrought to you byCOREprovided by Texas A&M UniversityA QUASI-DYNAMIC HVAC AND BUILDING SIMULATIONMETHODOLOGYA ThesisbyCLINTON PAUL DAVISSubmitted to the Office of Graduate Studies ofTexas A&M Universityin partial fulfillment of the requirements for the degree ofMASTER OF SCIENCEMay 2012Major Subject: Electrical Engineering

A QUASI-DYNAMIC HVAC AND BUILDING SIMULATIONMETHODOLOGYA ThesisbyCLINTON PAUL DAVISSubmitted to the Office of Graduate Studies ofTexas A&M Universityin partial fulfillment of the requirements for the degree ofMASTER OF SCIENCEApproved by:Co-Chairs of Committee, Charles CulpPrasad EnjetiCommittee Members,Don RussellPierce CantrellHead of Department,Costas GeorghiadesMay 2012Major Subject: Electrical Engineering

iiiABSTRACTA Quasi-Dynamic HVAC and Building Simulation Methodology. (May 2012)Clinton Paul Davis, B.S., Texas A&M UniversityCo-Chairs of Advisory Committee: Dr. Charles CulpDr. Prasad EnjetiThis thesis introduces a quasi-dynamic building simulation methodology whichcomplements existing building simulators by allowing transient models of HVAC(heating, ventilating and air-conditioning) systems to be created in an analogous way totheir design and simulated in a computationally efficient manner. The methodologyrepresents a system as interconnected, object-oriented sub-models known as components.Fluids and their local properties are modeled using discrete, incompressible objectsknown as packets. System wide pressure and flow rates are modeled similar to electricalcircuit models. Transferring packets between components emulates fluid flow, while thesystem wide fluid circuit formed by the components' interconnections determines systemwide pressures and flow rates.A tool named PAQS, after the PAacketized Quasi-dynamic Simulationmethodology, was built to demonstrate the described methodology. Validation tests ofPAQS found that its steady state energy use predictions differed less than 3% from acomparable steady state model. PAQS was also able to correctly model the transientbehavior of a dynamic linear analytical system.

ivTo my mother Melanie.

vACKNOWLEDGEMENTSI would like to acknowledge Dr. Culp for all of his help, guidance and insight. Iwould also like to thank Christina Chea for her help with editing. Finally, I would liketo thank my family for their support over the years.

viTABLE OF CONTENTSPageABSTRACT .iiiACKNOWLEDGEMENTS . vTABLE OF CONTENTS . viLIST OF FIGURES.viiiLIST OF TABLES . xi1.INTRODUCTION. 11.11.22.LITERATURE REVIEW. 42.12.22.32.42.52.63.HVAC Control Methodologies . 4HVAC Control System Design . 5Building Simulation Methodologies and Programs . 6Fluid Network Modeling . 9Modeling HVAC Components . 11Simulation Methodology Validation . 15HIGH LEVEL ALGORITHM DESIGN . 163.13.24.Background and Problem Statement . 1Objectives . 2Program Division . 16Simulations . 20LOW LEVEL ALGORITHM DESIGN . 244.14.24.3Component Outputs and Inputs . 24Modeling Fluids . 27Fluid Circuit Based Pressure and Flow Calculations . 42

viiPage5.VALIDATION AND TESTING. 585.15.25.36.Unit Testing . 58Steady State System Test . 59Dynamic System Test. 64CONCLUSION . 716.16.2Results Summary. 71Future Work . 72REFERENCES . 73VITA . 81

viiiLIST OF FIGURESPageFigure 3-1 – The Abstraction Layers of PAQS . 17Figure 3-2 – A Simple HVAC System . 19Figure 3-3 – Component Interconnections of a Simple HVAC System . 19Figure 3-4 - Pseudocode for the Main Simulation Algorithm . 22Figure 3-5 – Pseudocode for the Settle Phase of the Main Simulation Algorithm . 22Figure 3-6 – Pseudocode for the Progress Phase of the Main Simulation Algorithm . 23Figure 4-1 – Control Connections for a Simple System . 26Figure 4-2 – A Section of Pipe with its Internal Packets . 28Figure 4-3 – A Packet Being Split Into Two . 28Figure 4-4 – Two Packets Being Merged Into One . 29Figure 4-5 – Worst Case Cooling Density Change of Air . 30Figure 4-6 – Worst Case Heating Density Change of Air . 30Figure 4-7 – Air Density Changes Past a Heating Coil. 31Figure 4-8 – Heating Coil Density Changes Using Packets . 31Figure 4-9 – The Ports of a Cooling Coil . 32Figure 4-10 – A Directly Triggered Pipe . 34Figure 4-11 – A Packet Being Sent Into a Pipe . 34Figure 4-12 – A Pipe After Packet Propagation . 34Figure 4-13 – Invalid and Valid Output Packet Sizes for a Duct . 35

ixFigure 4-14 – A Chilled Water Loop with Internally Held Packets . 36Figure 4-15 – A Pipe Before Accepting a Packet . 37Figure 4-16 – Fluid Pushed Out of a Pipe . 37Figure 4-17 – A Pipe’s Output Packet . 37Figure 4-18 – A Large Air Packet Being Sent Into a Smaller Duct . 38Figure 4-19 – The Fluid Output of a Duct . 38Figure 4-20 – The Packet Sent to Another Component . 38Figure 4-21 – The Port Connections of an Air-Water Heat Exchanger . 39Figure 4-22 – An AHU Component with Different Flow Directions . 39Figure 4-23 – Flow Directions in an HVAC System . 40Figure 4-24 – A Single Zoned Building Represented as a Fluid Circuit . 43Figure 4-25 – A Dual Duct Mixing Box’s Fluid Circuit . 44Figure 4-26 – A 60 ft Long Duct with Known Port Pressures . 45Figure 4-27 – Pressure Along a Duct’s Length . 45Figure 4-28 – A Fluid Circuit Representation of a System . 47Figure 4-29 – Fluid Circuit Values on a Time Step . 47Figure 4-30 – Port Pressures on the Time Step . 48Figure 4-31 – A Valve’s Flow Rate . 48Figure 4-32 – A Typical Fluid Circuit Branch . 52Figure 4-33 – Combined Components of a Branch . 52Figure 4-34 – A System Represented as Circuit Branches . 53Figure 4-35 – An HVAC Cold Water Loop . 53

xFigure 4-36 – A Fluid Circuit of a Cold Water Loop . 54Figure 4-37 – A Branch Circuit of a Cold Water Loop. 54Figure 4-38 – Nodal Pressures of a Branch Circuit . 56Figure 4-39 – Calculating Port Pressures of a Circuit Branch . 56Figure 4-40 – Calculated Cold Water Loop Pressures . 57Figure 5-1 – A Single Duct Constant Air Volume System . 60Figure 5-2 – A Steady State System in PAQS . 62Figure 5-3 - Haberl vs. PAQS Cooling and Heating Usage Deviation . 63Figure 5-4 – Heat Transfer in a Package Unit AC System . 64Figure 5-5 – The Linear Analytical System in PAQS . 68Figure 5-6 – A System with an Overdamped Response . 70Figure 5-7 – A System with an Underdamped Response . 70

xiLIST OF TABLESPageTable 4-1 – Subtime Steps Required . 41Table 4-2 – Subtime Steps for Triggering Combinations . 42Table 4-3 – Values of Circuit Elements . 55Table 4-4 – Values of Branch Circuit Elements . 55Table 5-1 – Single Duct Constant Air Volume System Parameters . 61Table 5-2 – Nomenclature Used for the Analytical System . 65Table 5-3 – Values Used to Derive the System's Inputs . 69

11.1.1INTRODUCTIONBackground and Problem StatementDesigning successful controllers for HVAC systems that positively impactbuilding energy consumption and indoor air quality depends on the availability ofdynamic models that describe their key behaviors. Complexities in HVAC systems,such as distributed parameters and nonlinearities, make it difficult to obtain exactmathematical models (Tashtoush et al., 2005).The complexities of coupled interactions in buildings require computersimulations to predict energy consumption and system performance (CERL, 1999).Programs such as SPARK (LBNL, 2003), TRNSYS (Klein et al., 1976), andHVACSIM (Park et al., 1985) perform dynamic simulations of buildings using amodular equation solver approach. Modules generally consist of linear or nonlinearalgebraic or differential equations with the equation input and output variables serving asthe module's inputs and outputs. A module's connections with other modules determinethe overall system design. When performing a simulation, a computational equationsolver simplifies a system's equations and determines a solution scheme.While these methodologies can accurately model a building's dynamicperformance, improvements can be made. Iterative calculations involved in nonlinearequation solving take up considerable computational effort. Eliminating orThis thesis follows the style of ASHRAE Transactions.

2reducing these iterations would allow simulations to be performed more quickly.HVAC designers also view systems as linked objects that carry fluid and control signals,not as coupled sets of equations. Presenting simulations in this manner would makethem more intuitive to industrial users. Overall, developing a dynamic HVACsimulation tool that focuses on computational efficiency and usability would benefitHVAC control system designers and engineers to better understand the dynamics thatare involved with their systems.1.2ObjectivesThe concept of a new HVAC and building simulation methodology wasdeveloped and prototyped as part of the research involved with this thesis. Thismethodology simulates dynamic behavior over time using a series of steady statesimulations. Each steady state simulation covers a discrete period of time. Themethodology’s quasi-dynamic nature varies operating conditions between time stepswhile modeling a system as steady state on a particular time step.The design of this methodology focuses on three major areas. First, its structuremust be intuitive so those people familiar with HVAC control system design can createand simulate dynamic HVAC system models. Second, coupling between systemcomponents needs to be defined on the user’s level so that novel HVAC system modelscan be created from a library of standard components. Finally, it must becomputationally efficient so that it is practical for use in the work flow of HVAC controlsystem designers.

3After developing the simulation methodology, a prototype graphical userinterface and component library were designed and implemented in order to test themethodology and demonstrate its flexibility. The performances of overall simulationswere compared against previously validated steady state and analytical dynamic models,to ensure that the simulations worked as designed. Other tests were also performed toensure that components could be linked in general configurations and simulatedcorrectly.

42.2.12.1.1LITERATURE REVIEWHVAC Control MethodologiesComplete HVAC SystemsHVAC systems typically contain multiple interacting control loops that performseveral functions, such as maintaining the water level in a boiler, controlling the outputwater temperature of a chiller, and throttling a fan. As a result of the numerous systemconfigurations available, the method of control used in a particular control systemdepends heavily on the engineers involved in the design (Levenhagen, 1999).2.1.2Feedback Control MethodsCurrent control methods include analog electromechanical (pneumatic), analogelectronic, and digital. Pneumatic HVAC controls use compressed air to perform thecontrol logic and mechanical actuation of HVAC systems. Sensors output a pressuredifferential, and fluid logic circuits control pneumatic actuators (Wilson et al., 1965).Analog electronic control systems operate similarly to pneumatic systems using analogelectronic sensors, analog computers, and electromechanical actuators (Gupton, 2002).Direct digital control, or DDC, makes all control decisions and performs allcommunication with digital circuits (Newman, 1994).2.1.3HVAC Control AlgorithmsHVAC systems make extensive use of two term proportional-integral, or PIcontrollers, and occasionally use three term proportional-integral-derivative, or PID,controllers. The output of these types of controllers consists of the weighted sum of the

5input error, the time integral of the input error, and the time derivative of the input error.Terms that stabilize a system may be set using empirical methods such as the ZieglerNichols method or by individual hand tuning (Letherman, 1981).Self-tuning controllers in HVAC systems come in two types: autotuning andadaptive. Autotuning software automates the controller tuning procedure by exciting aresponse from a control loop, and calculating the control parameters from this response.Adaptive controls change their control strategy based on a well-known, but slowlyvarying operating condition (CIBSE, 2000).2.2HVAC Control System DesignIn the current literature regarding application of modern control techniques toHVAC, most use state space analytical representations of HVAC systems for controlsystem synthesis and analyses. For example, Argüello-Serrano and Vélez-Reyes (1995)developed a state observer for estimating a building’s thermal loads. Also, Moshiri andRashidi (2004) created an adaptive proportional-integral-derivative controller for HVACsystems using fuzzy logic.State space HVAC system models allow control system designers to analyticallyanalyze the effects of different changes in a system. However, they neglect manyparameters. To test and validate control system designs made using a simplified model,Anderson et al. (2007) constructed a physical HVAC system. The dynamic simulationtool Simulink (MathWorks, 2011) was used to make all feedback control decisions thatallowed them to configure and test new control methods entirely within software.Hepworth and Dexter (1994) tested and validated their control system using an actual

6building. They combined a traditional proportional-integral controller with a neuralnetwork for modeling nonlinear HVAC plant characteristics.2.32.3.1Building Simulation Methodologies and ProgramsNoncomputerized Building SimulationsDegree-day (DD) methods assume that a building’s daily heating or coolingusage varies proportionally to the difference between the daily mean outdoortemperature and a certain balance point temperature, typically 65 F (Knebel, 1983).Although DD methods provide rough estimates of a building’s energy usage, they lackthe capability to perform a detailed analysis of a building’s physics (ASHRAE, 2001).Bin methods, such as the Modified Bin method (Knebel, 1983), perform steadystate building simulations under a variety of weather conditions, and estimate energyusage for a given climate based on how often these weather conditions occur. They canaccount for individual system characteristics such as the part load performance ofHVAC equipment and heat pump systems.2.3.2Whole Building Energy AnalysisIn 1976, a collaboration of several national laboratories resulted in a buildingenergy analysis program known as DOE-2 (Birdsall et al., 1990). DOE-2 consists ofseveral subprograms that calculate different aspects of a simulation such as loads,HVAC system performance, plant performance, and costs. It simulates an entire year onan hourly basis, and includes dynamic building behavior that occurs on hourly timescales (York and Cappiello, 1981).

7In one analysis of DOE-2’s accuracy, the results came within 10% to 26% ofmeasured data values, within 1% to 30% of other energy simulation programs, andwithin 5% of analytical calculations (Haberl and Cho, 2004). In another analysis usingDOE-2, discrepancies of over 50% were found between simulated and measured datavalues. Differences in this case were found to be caused by the uncertainties inherent inbuilding energy simulation (Ahmad and Culp, 2006).Like DOE-2, BLAST consists of multiple programs that simulate a building onan hourly basis (CERL, 1999). However, BLAST uses a more advanced method basedon calculating zone loads (Strand et al., 2001). More computational effort is utilized fora more physically accurate calculation (Strand et al., 1999).EnergyPlus, released in 2001, serves as the successor to BLAST and DOE-2 byintegrating their most popular features and capabilities. It uses a modular architecturethat facilitates adding new features and links to other programs (Crawley et al., 1999). Itintegrates the calculation methodologies of previous programs, such as the thermal loadcalculations from a research version of BLAST, the daylighting calculation method ofDOE-2, the window modeling methods of WINDOW, and the interzonal airflowcalculation methods of COMIS (Strand et al., 1999). EnergyPlus structures simulationsaround components such as pipes, ducts, and chillers placed on abstract representationsof duct or pipe systems. During a simulation, all aspects of a building are simulated atonce, running an iterative algorithm until the entire system has converged on each timestep (Fischer et al., 1999).

82.3.3Equation Based Building SimulationsSPARK, (LBNL, 2003), represents buildings as interconnected modulescontaining nonlinear algebraic and differential equations. During a simulation, it solvesan entire decomposed equation set as a whole (Buhl et al., 1993). Libraries ofcomponents allow it to perform different tasks, such as hourly steady state energyanalyses and automated fault detection and diagnostics (Sowell and Moshier, 1995).TRNSYS, (McDowell et al., 2004), represents systems in the same way asSPARK. However, TRNSYS allows “black box” modules based on empirical datatables. Like SPARK, TRNSYS automatically generates calculation procedures for asystem (Klein et al., 1976). HVACSIM (Park et al., 1985) performs dynamicsimulations of buildings and HVAC systems using a hierarchical, modular approachbased on TRNSYS.2.3.4Interzonal Airflow AnalysisCOMIS, (Feustel, 1999), performs quasi-steady state pressure, airflow, andpollutant transport simulations of buildings. It represents buildings as connectedmodules that represent physical components such as zones, ducts, openings betweenzones, and fans. Users provide schedules that determine fan speeds and other systeminputs for a simulation.CONTAM, (Walton and Dols, 2005), performs the same type of airflowmodeling as COMIS. However, it uses a transient airflow model whenever non-flowprocesses like humidity removal exist. For modeling 1D convection at high air flowrates, CONTAM divides the air within a duct or zone into constant volume cells.

9Adding and removing cells from opposite ends of a duct or zone on each time stepmodels 1D convection exactly.2.3.5Computational Fluid DynamicsComputational fluid dynamics, or CFD, numerically solves the Navier-Stokesequations for fluid flow. Models that approximate turbulent flow must be used underturbulent flow conditions. Using CFD in HVAC requires an in-depth knowledge of fluidflow and the approximations used in the CFD modeling (Chen, 1997), due to thecomplexities of the numerical algorithms.2.3.6Coupling Different Building Analysis MethodologiesDifferent methodologies used in the analysis of buildings can be coupled togetherby (1) expanding the capabilities of existing software or by (2) facilitatingcommunication between software tools (Djunaedy et al., 2005). Coupling buildingsimulation tools that focus on different domains, such as fluid flow and heat transfer,combines the sophisticated methods used in each tool. Expanding current softwareeliminates the need for communication between two different programs, but may requirerewriting major portions of a program’s source code.2.4Fluid Network ModelingFluid networks, such as pipe and duct systems in HVAC, model fluid flow as acircuit with fluid traveling between sets of nodes. This section reviews fluid networkmodeling methodologies outside the area of HVAC that can be applied to it.

102.4.1Blood Flow SimulationOther areas of science develop fluid network models in ways that apply to otherfields. To model the human circulatory system, Migliavacca et al. and Westerhof et al.(2000; 1968) used passive electrical components as analogies to blood flow and bloodvessel behavior. This method derives from the Navier-Stokes equations and Hooke’slaw. By comparing to electrical analogies, voltage serves as an analog to pressure,current serves as an analog to flow, resistors serve as an analog to arterial flow resistance,inductors serve as an analog to fluid inertia, and capacitors serve as analogs to arterialelasticity.2.4.2Municipal Water DistributionThe numerous branches and segments found in HVAC ductwork and pipeworkresemble the structures found in municipal water distribution systems. In such networks,nonlinear equations relate the flows through each network element, and the head dropbetween the network nodes. When treated as steady state systems, the Newton-Raphsonmethod can be used to calculate all head losses and flow rates throughout the networkfor a single instant in time. Linearizations of these equations can be used in a dynamicsimulation, provided that changes in operating conditions remain small (Coulbeck, 1980).Filion and Karney (2003) found that simulations of water distribution networkshave many potential sources of error. Generating a system model from measured datahas the greatest error potential, due to the difficulty of testing and modeling each sectionof a network. Other fundamental sources of error exist as well. For instance, the

11estimates of the Darcy-Weisbach and Hazen-Williams friction models become unreliableunder transient conditions.2.5Modeling HVAC ComponentsHVAC system models combine the models of various subcomponents to createan entire system. This section reviews the modeling methodologies currently used onthese components. This ensures that any system-wide methodologies developed permitaccurate models of all parts of a system.2.5.1Boilers and ChillersWhile boilers often have complex internal controls, a single value representingtheir overall efficiency can represent the effects numerous model-specific parameters.Coil models can be used to simulate their dynamic performance (Garcia-Borras, 1983).Creating dynamic models of vapor compression chillers requires modeling all internalcomponents and control algorithms in order to model effects such as start-up, feedbackcontrol, and shut-down (Bendapudi et al., 2005).2.5.2Cooling and Heating CoilsThree main types of dynamic coil models exist: lumped capacitance models thatuse transfer functions to model heat conduction, dynamic models considering spatialvariation, and empirical models based on heat exchanger effectiveness (Yu et al., 2005).In order to develop an optimized control for cooling coils, Wang et al. (2004) developeda simplified steady state model with empirically chosen coefficients. Also, Xu et al.(2006) developed a dynamic cooling coil model for use in SPARK that divides a coil

12into twenty independent and discrete tube sections. This allows the modeling ofpartially wet coil conditions.2.5.3Cooling TowersCooling tower analysis applies energy and mass balance equations to theevaporation, heat flow, and mass flow that occurs within them. Commonly usedanalysis methods make several simplifying assumptions, such as neglecting heat transferwithin the circulating water (Webb, 1984).2.5.4Dampers and ValvesOne dimensional (fluid circuit) damper and valve flow modes use empiricallyderived function of a device’s stroke position and the ratio of its pressure drop versus thesystem’s total pressure drop. Manufacturers typically publish graphs of these flowcharacteristics. Their actuation method limits their dynamic response (Ward-Smith,1980).2.5.5Ducts and PipesAir in ducts loses energy through heat and air losses. As conditioned air passesthrough them, it can lose 10% to 40% of its cooling capacity and 0.5 C to 6 C throughheat losses with the surrounding air (Fisk et al., 2000). R-values can be used to quantifythe magnitude of such losses (Griffiths and Zuluaga, 2004). Duct air leakage typicallyaccounts for approximately 25% of the flow through fans. In commercial duct systems,most of this leakage typically occurs at joints versus the seams. Leakage rates typicallyvary as an exponential function of the static pressure difference at the duct’s surface(Aydin and Ozerdem, 2006).

13Compared to ducts, pipes typically lose more energy through radiation. Also,pressure losses due to valves and fittings usually exceed that of straight runs. Pipes canalso experience water hammer when water abruptly stops flowing. Over time, cavitationand other factors cause pipe system corrosion, changing the relationships betweenpressure and flow (Vedavarz et al., 2007).Modeling flows in pipes and ducts in multiple dimensions account for fluidchanges along the cross section of a flow (Ward-Smith, 1980). In a one-dimensionalanalysis, empirical formulas relate system pressure losses to its physical properties,internal fluids, and internal volume flow rate under ideal flo

Overall, developing a dynamic HVAC simulation tool that focuses on computational efficiency and usability would benefit HVAC control system designers and engineers to better understand the dynamics that are involved with their systems. 1.2 Objectives The concept of a new HVAC and building simulation methodology was