This item was submitted to Loughborough's Research Repository by the author. Items in Figshare are protected by copyright, with all rights reserved, unless otherwise indicated.

Teaching building performance simulation: ever done an autopsy?Teaching building performance simulation: ever done an autopsy?

PLEASE CITE THE PUBLISHED VERSION

http://www.bso16.org/

PUBLISHER

IBPSA

VERSION

AM (Accepted Manuscript)

PUBLISHER STATEMENT

This work is made available according to the conditions of the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) licence. Full details of this licence are available at:https://creativecommons.org/licenses/by-nc-nd/4.0/

LICENCE

CC BY-NC-ND 4.0

REPOSITORY RECORD

Beausoleil-Morrison, Ian, and Christina J. Hopfe. 2019. “Teaching Building Performance Simulation: EverDone an Autopsy?”. figshare. https://hdl.handle.net/2134/21257.

TEACHING BUILDING PERFORMANCE SIMULATION:EVER DONE AN AUTOPSY?

Ian Beausoleil-Morrison1 and Christina J Hopfe21 Faculty of Engineering and Design, Carleton University, Ottawa, Canada

2School of Civil and Building Engineering, Loughborough University, Loughborough, UK

ABSTRACTIn previous papers we have presented a continuouslearning cycle that includes exposure to theories andthe application of tools from the start for effectivelyteaching BPS and we have described the course wehave developed based upon this cycle. The importantrole played by the simulation autopsy in this cycle isthe focus of the current paper. This is accomplishedby examining the teaching methods we use for 2 ofour course’s 15 topics: determining the distribution ofsolar heat gains to internal building surfaces, and pre-dicting solar irradiance on external building surfaces.

INTRODUCTIONNow in its fifth decade, the building performance sim-ulation (BPS) field offers users a myriad of tools, sometargeted at practitioners, others aimed at researchersand expert users. Many have modern user interfacesand integrated performance overviews which allowusers to quickly ascend the learning curve to operatethe tools in order to produce simulation predictionswith minimal training. User manuals typically treatapplications but do not discuss default settings anddata, and the underlying modelling principles uponwhich the tools are based.Numerous books that treat the theoretical basis of BPSare available, for example: Clarke (2001), Under-wood and Yik (2004), Hensen and Lamberts (2011),and Peuportier (2016). But much of the theoreticalmaterial elucidated in such publications is not beinglearnt or understood by tool users. In an earlier paper(Beausoleil-Morrison and Hopfe, 2015), we presenteda series of observations on the current state of the fieldand argued the need for a different approach for teach-ing BPS, one that exposes users to the theoretical ba-sis.Various approaches have been used for teaching BPS.For example, Jankovic (2012), Struck et al. (2009),and Kumaraswamy and Wilde (2015) use a design/project focused approach. Gaming methods (Rein-hart et al., 2012) and assignment/ topic focused ap-proaches (Bernier et al., 2016; IBPSA-USA) have alsobeen used.We have taken a different approach which involveshaving students actively experiment with BPS tools tosupport the theoretical study of modelling and sim-ulation theory through a continuous learning cycle.The pedagogical basis of our approach as well asa description of the course we have developed have

been treated in earlier papers (Beausoleil-Morrisonand Hopfe, 2015, 2016).An important part of our BPS learning cycle is the sim-ulation autopsy, which is the subject of the current pa-per. The next section describes the simulation autopsy.We then present how such autopsies are key to the fulllearning cycle by presenting two of the topics of thecourse we have developed: modelling solar gains tointernal building surfaces, and modelling solar irradi-ance incident on exterior building surfaces.

TEACHING WITH AN AUTOPSYWe introduced the BPS-specific recursive learning cy-cle in Beausoleil-Morrison and Hopfe (2015) compris-ing four modes of learning: abstract conceptualization(AC), active experimentation (AE), concrete experi-ence (CE), and reflective observation (RO). The coursewe have developed (Beausoleil-Morrison and Hopfe,2016) consists of 15 iterations of this learning cycle.Each iteration treats a specific topic (two are treatedin this paper) and includes lectures on theory, assignedreadings, a simulation assignment, and a simulationautopsy. Active participation of the students through-out the cycle is critical.In Beausoleil-Morrison and Hopfe (2015) we ob-served that simulation predictions are often insuffi-ciently scrutinized by users and that users often placetoo much faith in their simulation tools; evidence gath-ered through initial testing of teaching methods wasprovided to substantiate these observations. We be-lieve that students need to be taught how to criticallyexamine their simulation results to identify and diag-nose sources of input error in order for them to developa necessary degree of skepticism, and the skills to di-agnose, repair, and reduce errors.This is why we introduced the concept of the simula-tion autopsy, a working session in which the studentsand the instructor collectively examine the results ofthe simulation exercise and dissect simulation inputfiles to diagnose reasons for predictive disagreement.We use this as a vehicle for helping students developcritical thinking skills for building upon the AE modeof learning and for relating to the theory acquired dur-ing the AC mode. The simulation autopsy takes placein the CE mode and then further blends into the ROmode of the learning cycle.The concept is simple: the students submit their resultsfor a simulation exercise, we compile them, and thenwe collectively analyze and scrutinize the results at the

Figure 1: Ambient air temperature (0C) from the weather file (left) against the results of the maximum zone airtemperature (in 0C, middle) and the annual space heating load (in GJ, right)

next class. The analysis phase often leads into groupdiscussions relating the relevant theory covered in thelectures and assigned readings to the impact upon sim-ulation predictions. As well, the simulation input filesprepared by individual students are sometimes collec-tively examined to help diagnose causes of predictivedisagreements and to identify potential input errors.This reflective observation and diagnostic investiga-tion is what connects the experience to the theory. Asstated by Dewey (1933), reflective thought is an “ac-tive persistent and careful consideration of (..) sup-posed form of knowledge in the light of the groundsthat support it and further conclusion to which ittends”. We argue that in most literature and courseson BPS this final—and crucial—stage that motivatesreflective thought is omitted, and this is key to trans-forming the student from a learner to a reflectivelearner.We pose a number of questions to the students duringthe simulation autopsies to facilitate the reflection pro-cess, such as:

• Sanity check: How do we expect the building toperform?

• Can you trust the results?• What do the results mean?• What evidence supports your answers to the

above questions?To illustrate the above in more detail we return tothe base case assignment introduced in Beausoleil-Morrison and Hopfe (2016) that we hand out to the stu-dents at the beginning of the course in order for themto develop a basic familiarity with the operation of thechosen BPS tools, and to develop an appreciation ofthe types of data required to describe a simple singlezone building.In a recent offering of the BPS course at Carleton Uni-versity, each student conducted this exercise indepen-dently using both EnergyPlus and ESP-r and submit-ted the results in advance of the course’s first simula-tion autopsy. All students managed to produce results

with both ESP-r and EnergyPlus only three weeks af-ter their first exposure to these tools.Figure 1 shows a compilation of some of the resultsthat the 9 students submitted and which were collec-tively examined during the simulation autopsy.On a first glance, some of the maximum zone airtemperature predictions vary significantly: Student I’sESP-r result is 120C higher than Student B’s Energy-Plus result. Whilst the majority of all students agreeon timing of this peak temperature (mid-afternoon onAugust 19), three students notice a different day us-ing Energy Plus, whilst three other students notice adifferent day with ESP-r. There are also significantdifferences in the predicted annual space heating load:Student C’s and I’s EnergyPlus results are significantlyhigher than the others.The group discussed these differences and hypothe-sized explanations. Following this, some of the Ener-gyPlus and ESP-r files prepared by the students wereprojected on a screen, the inputs examined, and diag-nostic techniques explained for searching for potentialinput errors. A number of discoveries were made dur-ing this portion of the simulation autopsy:

• The annual space heating load predicted by Stu-dent C using EnergyPlus was quite high. Thereason for this was that the two south facingwindows were placed on the north face of thebuilding instead.

• In student I’s EnergyPlus model the constantsensible internal heat gain of 200W was spec-ified as latent; this increased the annual spaceheating significantly.

• Student H’s ESP-r model shows that the annualspace heating load is quite high despite the max-imum zone air temperature and peak heatingload projections being in line with the others.Furthermore, the peak zone air temperature oc-curs a month earlier than expected. It turned outthat this was due to using the wrong weather file.

• The cause of Student D’s EnergyPlus simulation

predicting a low annual space heating load wasfound to be an error (not trapped by EnergyPlus)in the vertex points defining the roof: in fact, itwas an incomplete surface and this resulted inan under-prediction of the heat loss through theroof.

• Student G’s ESP-r results generally look okayexcept that the results reported the peak heat-ing on the wrong day. This is because of usinga pre-conditioning period of only 1 day. (Thisprovided an excellent segue to return to the ear-lier presented theory about how numerical meth-ods require a pre-conditioning period to elimi-nate the impact of the initial condition.)

Needless to say that this list is not exhaustive but doesserve to illustrate the important lessons that can ariseduring the simulation autopsy. This concept will befurther explored in the next sections, which focus ontwo of the 15 iterations of the learning cycle used inour course: solar heat gains at internal surfaces andsolar irradiance at external surfaces.

SOLAR HEAT GAINS ATINTERNAL SURFACESThis section focuses on modelling the distribution ofsolar heat gains to internal building surfaces. The goalis for the students to understand the default and op-tional methods employed by BPS tools, and to dis-cern the sensitivity of simulation predictions to usersinputs and choices. We use a combination of lec-tures, assigned readings, video sequences, simulationexercises, and a simulation autopsy following the BPSlearning cycle (Beausoleil-Morrison and Hopfe, 2015,2016). The theory that is presented during the lecturesis explained in the following subsection.

Presentation of theory

This lecture builds upon earlier course material thatexplained how BPS tools form and solve energy bal-ances to determine a building’s thermal state as a func-tion of time. This described how energy balances areformed for each internal surface (wall, floor, ceiling),such as represented by the control volume (CV) delin-eated by the dashed line in Figure 2:

energystoredwithin

CV

=

{solar radiation

absorbed

}+

convectionfrom

indoor air

+

net radiationfrom other

internalsurfaces

+

net radiation

frominternalgains

+

net radiationfrom plantequipment

−

conductiontowardsoutside

(1)

Figure 2: Energy balance at internal building surfaces

The method universally employed by BPS tools to cal-culate the first term on the right side of Equation 1 foreach individual surface is presented:{

solar radiationabsorbed

}= α ·A · q′′solar,int (2)

Where α and A are the solar absorptivity (-) and area(m2) of the surface under consideration. The formeris a parameter supplied by the user while the latter isdetermined from the building’s geometrical inputs.q′′solar,int in Equation 2 is the solar irradiance inci-dent upon the internal surface under consideration(W/m2). This includes the beam and diffuse solarirradiance that is transmitted through windows andwhich directly strikes the surface, as well as the so-lar irradiance reflected off other surfaces towards thesurface under consideration.Explaining this theory makes the student aware thatonly three factors determine the prediction of solarheat gains to each internal surface: user-defined ge-ometry, user-prescribed α values, and the models usedto predict q′′solar,int. This prepares the student for thesimulation exercises which allow them to explore thedegree to which modelling decisions and user inputdata influence simulation predictions.

Simulation exercise

There are two steps to the simulation exercise, the firstof which explores the sensitivity of simulation predic-tions to α, often an uncertain parameter for the user toquantify.

Step 1: Perform a simulation using the base caseand extract the results for February 21 (a cooland sunny day). Create a graph that plots the so-lar radiation absorbed by the concrete floor (W)versus time for this day. Also plot the rate of heatinput from the convective HVAC system.

Perform a second simulation in which the so-lar absorptivity of the concrete floor is increased

from 0.6 to 0.9. Add the predictions from this sim-ulation to the above graph.

Did increasing the solar absorptivity of the con-crete floor have the expected impact upon theamount of solar radiation absorbed by the floorsurface? Explain why this significant change inthe floor’s solar absorptivity had minimal impactupon the quantity and timing of heat injection bythe HVAC system.

The second step of the simulation exercise motivatesthe students to understand the default method appliedby their chosen BPS tool for determining q′′solar,int aswell as to explore optional models.

Step 2: How did you configure your base case totreat the distribution of solar gains to the interiorsurfaces? If you did not use the default methodprovided by your BPS tool, then configure yourtool now to employ its default method. What is thedefault method employed by your BPS tool? Per-form a simulation, extract the results for Febru-ary 21, and create a graph as described in Step1.

Perform another simulation in which your BPStool applies its most detailed method for deter-mining the distribution of solar gains as a func-tion of time. Describe the algorithm employedby your BPS tool. Add the predictions from thissimulation to the above graph. Explain the differ-ences in the solar radiation absorbed by the floorover the course of this day.

What impact does this have upon the quantity andtiming of heat injection required from the HVACsystem? Describe a situation in which a greaterimpact would occur.

Figure 3: Step 1-Sample results produced by a studentfor the simulation exercise on solar heat gains at in-ternal surfaces.

Student results

An example of their results for Step 1 are illustratedin Figures 3 and 4, while Figure 5 combines some oftheir Step 2 results.

Figure 4: Step 1-Comparison of EnergyPlus and ESP-r predictions of solar energy absorbed by floor for thebase case.

Figure 5: Step 2-Comparison of tool default and op-tional methods for determining the distribution of so-lar gains to internal surfaces.

SOLAR IRRADIANCE ONEXTERIOR SURFACESThis section focuses on predicting solar irradiance onexterior building surfaces. The goal is for the studentsto understand the models that are employed for pre-dicting the solar irradiance on each building surface

based upon the data contained in weather files, to ap-preciate the inherent uncertainties in these models, andto develop an appreciation for the impact of user inputdata and choices on simulation predictions.The material treated during the lecture portion of thelearning cycle is outlined in the next subsection.

Presentation of theory

The lecture builds upon earlier material that explained:how solar gains to internal surfaces are determined (re-fer to the previous section); the types of solar datacontained in weather files; and external surface energybalances (similar in structure to the internal surface en-ergy balance given in Equation 1).The common method employed by BPS tools to calcu-late the solar energy absorbed by the exterior of eachsurface is presented:{

solar radiationabsorbed

}= α ·A · q′′solar,ext (3)

Where α and A are the solar absorptivity (-) and area(m2) of the external surface under consideration. Theformer is a parameter supplied by the user while thelatter is determined from the building’s geometrical in-puts.The lecture explains how the total solar irradianceon exterior surfaces (q′′solar,ext in W/m2) is predictedfrom the summation of three components:

q′′solar,ext =

directbeam

irrdiance

+

skydiffuseirrdiance

+

groundreflectedirrdiance

(4)

The methods used to determine each of the three ir-radiance components required by Equation 4 are ex-plained. This makes the students aware that deter-mining the direct beam component from the data inthe weather file depends only upon geometry, whereasthere is greater complexity in determining the skydiffuse and ground reflected components. Specifi-cally, they become aware of the empirical nature ofanisotropic sky diffuse models and the modelling op-tions available to the user, as well as the factors thatinfluence ground reflectivity. This prepares the stu-dent for the simulation exercise which allows them toexplore the degree to which such modelling decisionsand user input data influence simulation predictions.

Simulation exercise

There are four steps to the simulation exercise, the firstof which explores the sensitivity of simulation predic-tions to α, often an uncertain parameter for the user toquantify.

Step 1: Perform a simulation in which the solar

absorptivity of the asphalt roof membrane is re-duced from 0.85 to 0.6. What impact does thishave upon the annual space heating load? De-scribe a situation in which such a change wouldhave a greater impact.

The second step motivates the students to understandtheir chosen BPS tool’s default and optional methodsfor modelling sky diffuse radiation.

Step 2: What methods does your BPS tool pro-vide for predicting the distribution of diffuse skyirradiance? What is your tool’s default method?

Which method did you use for simulating the basecase? Perform one or more simulations using al-ternate methods offered by your BPS tool (e.g.isotropic sky model). What impact does this haveupon the annual space heating load? Commenton the uncertainty of simulation predictions re-sulting from the treatment of diffuse sky irradi-ance.

The third step is designed to make students aware ofthe importance of ground reflected irradiance, to ap-preciate the sensitivity of simulation predictions touser choices regarding ground reflectivity, and to ap-preciate the significance of accepting BPS tool de-faults.

Step 3: The base case specifications did not in-clude any information on ground albedo. Did youassume a constant value in your simulation and ifso what value did you assume? If you used yourBPS tool’s default treatment, then describe whatthis treatment is.

Perform a second simulation in which the groundalbedo has a constant value of 0.7. Create agraph that plots the solar irradiance incidentupon the exterior surface of the windows as afunction of time on February 21. Use this graphto contrast the predictions from the two simula-tions. Also contrast the predictions of the spaceheating load over this day.

Based upon these results, comment on the un-certainty introduced into simulation analyses dueto ground albedo. What are some situations inwhich this could be a significant factor in overalluncertainty?

The fourth step of the simulation exercise encouragesstudents to explore methods for describing solar shad-ing caused by surrounding objects and to develop amotivation for not ignoring such blockages.

Step 4: Now consider the solar shading causedby a tree located south of the building. Thetree has a diameter of 2 m, a height of 10 m,and is located 10 m from the south facade and

mid-distance between the two windows. Createa graph that plots the solar irradiance incidentupon the exterior surface of the windows as afunction of time on February 21. Use this graphto contrast the predictions from the base case.

Based upon these results, comment on the impor-tance of considering shading effects by surround-ing buildings, trees, and other objects. Whatare some situations in which the complexities ofshading objects (e.g. size and types of trees,changing leaf cover) could be a significant fac-tor in overall uncertainty?

Figure 6: Step 3-Sample results produced by a studentfor solar irradiance on exterior surfaces.

Figure 7: Step 4-Sample results produced by a stu-dent for solar irradiance on exterior surfaces with andwithout a tree located south of the building.

Student results

An example of the results generated by the CarletonUniversity students are presented in Figures 6 (Step 3)and 7 (Step 4).

CLOSING THE LEARNING CYCLE:THE SIMULATION AUTOPSYAs explained earlier, our learning cycle uses simula-tion autopsies as a vehicle for strengthening the con-crete experience and for transforming the student froma learner to a critical reflective learner. By focus-ing on the two course topics presented earlier—solarheat gains at internal surfaces and solar irradiance atexternal surfaces—this section describes how the au-topsy takes place during the CE mode and then furtherblends into the RO mode of the cycle.

Solar heat gains at internal surfaces

Step 1: Each student produced and submitted graphsas shown in Figure 3, and during the following classwe collectively examined these results and discussedtheir implications. The group observed that changingαfloor from 0.6 to 0.9 had the expected impact uponthe rate of solar energy absorbed by the floor (referto Equation 2 and compare the black circles and bluesquares in Figure 3).Given the influence of αfloor upon the solar energyabsorbed by the floor, most students expected this toaffect the zone energy balances (e.g. Equation 1) andconsequently the heat injection required by the HVACsystem. However, this hypothesis was disproved bythe simulation predictions. As can be seen in Figure3 (compare the red triangles to the purple pluses), thissubstantial change to αfloor had a negligible influenceon the required heating for this day. Indeed, the im-pact upon the annual space heating load was only 0.4%(with both ESP-r and EnergyPlus).The group discussed the reasons for this, referringback to the course’s earlier treatment of internal sur-face and zone air energy balances and came to the re-alization that—at least for well-insulated buildings—significant changes in αfloor would have only mini-mal impact upon the required heat injection. This isan important finding because in many situations BPStool users do not have access to precise α data, andassumptions or approximation are necessary; in thiscase, it is clear that significant time investment is notwarranted to determine these input values.During the collective examination of the Step 1 re-sults, a number of students noticed a significant dif-ference between the EnergyPlus and ESP-r predictions(see Figure 4), and this led to a discussion of the de-fault methods the two tools use to distribute solar gainsto internal surfaces. By default EnergyPlus assumesthat all solar beam irradiance entering a zone is inci-dent upon the floor; other surfaces receive diffuse ir-

radiance and irradiance reflected by the floor. By de-fault ESP-r distrubutes all solar irradiance entering azone diffusely on a area-weighted basis. (Many expe-rienced users of these tools are likely unaware of thesedefaults.) Although this topic had been covered duringthe lecture, for many students the implications of thesedefault methods only became clear during the simula-tion autopsy. This led to a discussion on situations inwhich these default methods may be inappropriate andcould cause important simulation errors.Step 2: Both EnergyPlus and ESP-r offer optionalmethods for predicting q′′solar,int. Through an explo-ration of the tool documentation and the demonstra-tions provided in class, the students discovered howto invoke the optional methods for more accuratelydetermining the distribution of solar gains by consid-ering building geometry and sun position as a func-tion of time. Step 2 of the exercise made them realizehow straightforward it is to invoke these more accu-rate models and the minimal impact they have uponcomputational resources.An example of the results produced by one student isillustrated in Figure 5, which contrasts the predictionsof solar gains absorbed by the floor for the tool defaultand optional methods. The blue circle symbols in thisfigure reproduce the results previously shown in Fig-ure 4 (tool default methods) whereas the results pre-dicted with ESP-r’s ray-tracing insolation algorithmand EnergyPlus’ direct geometrical method are illus-trated with red triangle symbols. Examination of theseresults led the students to realize that for this buildinggeometry and for this day of the year, that ESP-r’s de-fault method resulted in substantial errors, whereas theEnergyPlus default method provided a reasonable ap-proximation of the solar energy absorbed by the floor.Once again, an examination of the HVAC heat injec-tion predictions made the students realize that despitelarge differences between the ESP-r default and op-tional methods (refer to Figure 5), the impact uponthe functioning of the HVAC system was seen to benegligible. The group discussed these implicationsand a consensus was reached that it was worth invok-ing the optional detailed methods as they demandedlittle from the user and had minimal computationaldemands, although in many situations (such as withthis well-insulated building) the impacts upon someperformance metrics would be minimal. There wasalso a discussion about the limitations of the optionalmethods and the implications this had upon abstract-ing building geometry. For example, EnergyPlus em-ploys a direct geometrical method that is limited to so-called convex shaped floor plans, which places limi-tations how building geometry can be represented andsub-zoned.Perhaps the most important lesson the students learntfrom this exercise was that the user inputs for internal-surface material α values and choices on methods for

solar distribution models have a much smaller impactupon simulation predictions (for this building) thansome of the aspects explored in previous simulationexercises, such as the treatment of internal surface con-vection and internal heat and moisture gains.

Solar irradiance at exterior surfaces

Step 1: Given the insensitivity of simulation predic-tions to the solar absorptivity of the interior surface ofthe floor (refer to the previous simulation autopsy), thestudents were surprised to observe reducing the solarabsorptivity of the asphalt roof membrane increasedthe annual space heating load by 2.5 to 3.5%. Thegroup discussed the reasons for this, referring back tothe theory of the internal and external surface energybalances, and how reflected irradiance was treated.The conclusion drawn was simulation predictions forbuildings like this are much more sensitive to the solarabsorptivity of exterior surfaces than to that for inter-nal surfaces, and that users should invest more time insecuring accurate data for the former.Step 2: Both the BPS tools used by the students em-ploy the same model (Perez et al., 1990) by defaultfor predicting anisotropic sky diffuse irradiance. Con-ducting Step 2 of the simulation exercise made the stu-dents realize that EnergyPlus does not provide any op-tional methods, while ESP-r allows the user to selectalternate anistropic models or to treat the sky domeas isotropic. Each student chose one of these op-tional methods for conducting an ESP-r simulationand compared their results during the simulation au-topsy. Through this, the group realized that the im-pact of model choice on the annual space heating loadwas significant: differences as much as 11% were ob-served.Referring back to the theory presented in the lec-tures on sky models, the students realized why treatingthe sky as isotrpic—specifically, neglecting brighten-ing around the horizon—may lead to significant dif-ferences in the prediction of total irradiance incidentupon windows (and therefore solar gains to the build-ing). This observation also motivated them to reexam-ine the structure of the Perez et al. (1990) anisotropicsky model and made them realize there must be un-certainty associated with its model form and empiricalconstants, and this could propagate into significant un-certainty in BPS predictions.Step 3: An example of the Step 3 results producedby one student is illustrated in Figure 6, which con-trasts the predictions of total solar irradiance incidentupon the south-facing windows (q′′solar,ext in Equation4). Each of the students produced and submitted suchgraphs which were collectively examined during thesimulation autopsy. The group observed that chang-ing the ground reflectivity from 0.2 (a common de-fault value in BPS tools and appropriate for dry bareground) to 0.7 (a value more representative of fresh

snow) had a significant impact (compare the blue cir-cles to the red triangles in Figure 6). The impact onthe annual space heating load was seen to be approxi-mately 18%.Although this exercise may represent the extremes ofthe bounds on ground reflectivity, it served to makethe students aware that the reliance on BPS tool de-fault values may lead to significant errors, particularlyin locations where snow cover is frequent.Step 4: By conducting Step 4 of the simulation ex-ercise the students became aware of the facilities of-fered by EnergyPlus and ESP-r for representing exter-nal shading objects, including the data required fromthe user and the level of effort involved. An exam-ple of the results produced by one student is illustratedin Figure 7, which contrasts q′′solar,ext on two south-facing windows.Although this exercise prescribed an idealized shapefor the tree, it motivated the students to explore theshading facilities and to realize the complexities in-volved in representing external shading objects, suchas geometry and how opacity changes with the sea-sons. The group had a lively discussion on how mucheffort BPS tool users should invest in this task and howmuch uncertainty it introduces into an analysis if suchshading is ignored (a common practice): in this casethe tree increased the annual space heating load by ap-proximately 9%.

CONCLUDING REMARKSThe simulation autopsy—a working session in whichthe students and the instructor collectively examinethe results of simulation exercises and dissect sim-ulation input files to diagnose reasons for predictivedisagreement—plays a crucial role in our BPS learn-ing cycle. We use this as a vehicle for helping studentsdevelop critical thinking skills and to connect their ac-tive experimentation with BPS tools with the theorythat is taught through lectures and assigned readings.The organization and value of the simulation autopsywas illustrated in this paper by focusing on 2 of the 15topics of our BPS course: solar heat gains at internalsurfaces and solar irradiance at external surfaces. Asuccinct summary of the theory we present on thesetopics is provided and the simulation exercises we as-sign the students were presented. An example of theresults generated by the students was provided, andthen the simulation autopsy sessions described. Al-though space limitations prevented an exhaustive de-scription, we attempted to provide here some glimpsesinto the types of learning that were accomplished dur-ing these sessions.We have observed that for many students the implica-tions of BPS tool default models became clear onlyafter the simulation autopsy and that these sessionsare critical for consolidating the teachings of theory.Besides a significant reduction in user errors, the au-

topsies always led to discussions on whether certainmethods may be appropriate or not.The student experience also reflected our observa-tions. Feedback surveys were administered frequentlyto assess the effectiveness of the teaching methods atachieving the stated intended learning objectives foreach topic. The students consistently rated the simula-tion autopsies as the most helpful method for support-ing their learning (conducting the simulation exercisescame a close second).

REFERENCESBeausoleil-Morrison, I. and Hopfe, C. J. 2015. Teach-

ing building performance simulation through a con-tinuous learning cycle. In Proc. Building Simulation2015, pages 2757–2764, Hyderabad, India.

Beausoleil-Morrison, I. and Hopfe, C. J. 2016. De-veloping and testing a new course for teaching thefundamentals of building performance simulation.In Proc. eSim 2016, Hamilton, Canada. Accepted.

Bernier, M., Kummert, M., Sansregret, S., Bourgeois,D., and Thevenard, D. 2016. Teaching a buildingsimulation course at the graduate level. In Proc.eSim 2016, Hamilton, Canada. Accepted.

Clarke, J. 2001. Energy Simulation in Building De-sign. Butterworth-Heinemann, UK, 2nd edition.

Dewey, J. 1933. How we think. Chicago, IL: Regency.Hensen, J. L. and Lamberts, R. 2011. Building Perfor-

mance Simulation for Design and Operation. Rout-ledge, UK.

IBPSA-USA. Building energy modelling trainingworkshop. http://www.ibpsa.us/resources. Ac-cessed: 2016-03-16.

Jankovic, L. 2012. Designing Zero Carbon BuildingsUsing Dynamic Simulation Methods. Routledge,UK.

Kumaraswamy, S. B. and Wilde, P. D. 2015. Simula-tion in education: Application in architectural tech-nology design projects. In Proc. Building Simula-tion 2015, pages 2773–2780, Hyderabad, India.

Perez, R., Ineichen, P., Seals, R., Michalsky, J., andStewart, R. 1990. Modeling daylight availabilityand irradiance components from direct and globalirradiance. Solar Energy, 44(5):271 – 289.

Peuportier, B. 2016. Energetique des batiments et sim-ulation thermique. Eyrolles, France.

Reinhart, C. F., Dogan, T., Ibarra, D., and Samuelson,H. W. 2012. Learning by playing – teaching energysimulation as a game. Journal of Building Perfor-mance Simulation, 5(6):359–368.

Struck, C., de Wilde, P. J., Hopfe, C. J., and Hensen,J. L. 2009. An investigation of the option spacein conceptual building design for advanced build-ing simulation. Advanced Engineering Informatics,23(4):386 – 395. Civil Engineering Informatics.

Underwood, C. P. and Yik, F. W. 2004. ModellingMethods for Energy in Buildings. Blackwell Pub-lishing, UK.