Longitudinal prediction of the operational energy use of buildings

Building and Environment 46 (2011) 1670e1680

Contents lists available at ScienceDirect

Building and Environment journal homepage: www.elsevier.com/locate/buildenv

Longitudinal prediction of the operational energy use of buildings Pieter de Wilde a, *, Wei Tian a, Godfried Augenbroe b a b

School of Architecture, Design and Environment, University of Plymouth, Plymouth, United Kingdom College of Architecture, Georgia Institute of Technology, Atlanta, USA

a r t i c l e i n f o

a b s t r a c t

Article history: Received 4 November 2010 Received in revised form 7 February 2011 Accepted 10 February 2011

Thus far most studies of operational energy use of buildings fail to take a longitudinal view, or in other words, do not take into account how operational energy use changes during the lifetime of a building. However, such a view is important when predicting the impact of climate change, or for long term energy accounting purposes. This article presents an approach to deliver a longitudinal prediction of operational energy use. The work is based on the review of deterioration in thermal performance, building maintenance effects, and future climate change. The key issues are to estimate the service life expectancy and thermal performance degradation of building components while building maintenance and changing weather conditions are considered at the same time. Two examples are presented to demonstrate the application of the deterministic and stochastic approaches, respectively. The work concludes that longitudinal prediction of operational energy use is feasible, but the prediction will depend largely on the availability of extensive and reliable monitoring data. This premise is not met in most current buildings. Ó 2011 Elsevier Ltd. All rights reserved.

Keywords: Longitudinal thermal performance Component degradation Climate change Building maintenance Operational energy use

1. Introduction The operational energy use of buildings (the energy used for heating, cooling, ventilation, provision of domestic hot water, lighting, and appliances) is the main focus of energy efficiency measures. More recently the building science community has started to balance this operational energy with the embodied energy invested in creating new buildings [1e3]. Thus far most assessments consider operational energy use to be constant over the building life cycle [1,2,4]. However, this is unlikely to hold true in real life [5]: the thermal performance of building components will be degrading over time[6], environmental conditions are subject to climate change, and internal loads will vary due to changes in building occupancy and when new technologies and appliances enter the market. As buildings become more efficient, these changes in energy performance during the building life cycle become more important factors in long term energy accounting methods. Furthermore, studies in for instance the field of adaptation of buildings to changing climate conditions require a longitudinal view of building performance rather than just an initial assessment. The longitudinal thermal performance of a building will be affected by organizational and physical change processes affecting building occupancy, facility use properties of building elements, climatic conditions, and building maintenance processes and policies.

* Corresponding author. Tel.: þ44 1752 586115; fax: þ44 1752 586003. E-mail address: [email protected] (P. de Wilde). 0360-1323/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.buildenv.2011.02.006

Information is becoming available to estimate the magnitude of these changes, for instance in the areas of properties of building components over time [7e9], maintenance processes [10], and climate change [11,12]. Other factors, like occupancy changes, are likely to remain hard to predict and will result in the need to accommodate uncertainties and risk when planning for the future. Note that embodied energy can be responsible for a considerable proportion (up to 45%) of the total energy for low energy buildings [3] but this topic is considered to be beyond the scope of this research. The purpose of this article is to propose a methodology for predicting the longitudinal thermal performance of buildings, specifically taking into account degradation in thermal performance of building components and changes in weather conditions. Firstly it reviews the current state-of-the-art in building component service life expectancy, degradation of building component thermal performance, building maintenance, and changing environmental conditions (climate change). Based on this review, a methodology is suggested to predict the likely change of thermal performance over the building life cycle. Two simple case studies are presented to demonstrate how the operational energy use is expected to change over its service life by using the deterministic and probabilistic methods, respectively.

2. Service life expectancy of building components This section reviews the service life expectancy of building components. For new or existing buildings, it is necessary to predict

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

the service life of building components to prepare long term maintenance and repair budgets in life cycle analysis [13]. Service life prediction of building components can be estimated based on two methods: deterministic and stochastic prediction [14]. It should be noted that replacement of building components may take place not only due to failure or anticipated failure but also due to obsolescence, change in building use, high maintenance costs or other reasons. Hence, the UK’s BCIS (Building Cost Information Service) [15] defines six forms of service life: physical, economic, functional, technological, social, and legal. These apply both to the whole building and to building components. 2.1. Deterministic and stochastic methods A commonly accepted deterministic method is the factor method [16], described in detail in ISO 15686. Here a reference service life is multiplied by six factors in the range from 0.9 to 1.1 (quality of components, design level, work execution level, indoor condition, outdoor environment, in-use conditions, and maintenance level). The choice for a value for each of these factors is highly dependent on the subjective judgement/expertise of the assessor. The reference service life of building components is available in the literature [15,17e20]. This method offers a practical assessment, but is limited in that it does not quantify service life as a stochastic quantity. In stochastic methods, the service life expectancy of building components is considered as a distribution rather than a point forecast. Markov models are commonly used for stochastic prediction [21]. Markov processes are classified according to the nature of time and the nature of the state space: continuous and discrete states. There is only limited research into continuous-state Markov processes. A discrete state Markov process is also called a Markov chain. In this chain, future chains are dependent only on the present state and independent of any previous states (these models are therefore called memory-less). Markov chains consist of an initial state distribution and a transition matrix. The transition matrix represents the probability of the process moving from state A to state B. For example, the condition of a roof can be regarded as A (excellent), B (satisfactory), C (unsatisfactory), and D (failing). Then the transition matrix for roof performance contains the probability of changeover of roof performance among the four conditions (A, B, C, and D). This matrix of state transition probabilities is usually obtained from a vast amount of historical maintenance data [22,23]. The quality and quantity of data is in fact a major challenge to the applicability of Markov chains. The application of Markov chains in construction has been demonstrated by Zhang et al. [24]. 2.2. Engineering design methods The above two methods have apparent drawbacks. The factor method only gives a single value of service life without considering the variability of actual conditions, while the probabilistic method is likely to be too complicated to be used for most applications. Moser and Edvardsen [25] have proposed engineering design methods by improving the factor method, in which the factors described in Section 2.1 are redefined as density distributions instead of a single figure. The equation for the factor method in ISO 15686 can be used for standard cases while modified equations may need to be established in some special cases. Then the probability distribution functions of these factors are determined by expert opinions, such as Delphi method. The results from this method would be expressed as a distribution to reflect the variability of the processes. Examples of application of this method can be found in [25].

1671

3. Degradation of building components There is very limited quantitative data available on the degradation of the thermal performance of buildings over time. Changes in building thermal performance due to degradation of building components and available deterioration models are described in this section. Building degradation is closely related to building maintenance, which will be described in Section 4.2. 3.1. Review of deterioration of building components One way of looking at the elements of a building is to separate them into ‘building components’ and ‘engineering components’ [21]. Here the building components include building envelope, like walls, roofs, windows and doors. The engineering components include the mechanical equipment, like fans, pumps, boilers, and chillers. The forms of ageing of these building elements may exhibit a combination of linear, abrupt, and exponential characteristics, depending on the intrinsic properties, exposure and use circumstances. As an example of degradation of a building component, the insulation material of the envelope may gradually degrade after installation and, consequently, the wall heat resistance may decline with time. The overall insulation benefit during the building life is determined by long term thermal resistance, and less by its initial value at delivery, as the building envelope has a relatively long service life. In general a building envelope should have a service life of at least 25 years without major maintenance and may need refurbishment to function for another 20e25 years [26]. In more detail, the long term thermal performance of closed-cell foam insulation can be calculated from two test methods: CAN/ULC-S770 in Canada, and ASTM C1303 in the USA [27]. Both methods calculate the long term thermal resistance values (LTTR) in the form of fiveyear thermal conductivity values. Note that according to Graham [28] even when using such methods, for extruded polystyrene (XPS) insulation the R-value obtained from CAN/ULC-S770 may still be overestimated. The distributed parameter continuum model (DIPAC) developed by IRC/NRC (the Institute for Research in Construction/National Research Council of Canada) is a tool used to evaluate the LTTR for foam insulation material [29]. Agesim [27], developed by Huntsman Polyurethanes, is another tool to predict the thermal ageing for XPS and polyiso boards. Based on the research of PU-EUROPE [30], the thermal conductivity for rigid polyurethane foam (PUR/PIR) becomes stable after three years, and the thicker the insulation material the lower the expectation is about the long term thermal conductivity. But again, Graham [31] indicates that actual R-values for polyisocyanurate board insulation are lower than the LTTR from manufacturers in northern climates of Canada. As an example of degradation of an engineering component, the efficiencies of HVAC equipment may be reduced due to normal wear and tear or other reasons. The thermal performance and efficiency of boilers, chillers and heat exchangers can stem from reduced heat transfer because of low refrigerant charge, fouling, or burning problems [32]. Bannai et al. [33] found that the COP (coefficient of performance) of a turbo chiller is reduced by 2.53% annually. However, most of this performance deterioration (around 68%) is due to the contamination of the tube and can be recovered by using chemical cleaning. In a report from NREL (National Renewable Energy Laboratory, USA), Griffith et al. [34] assume that the chiller COP and boiler efficiency degrade at a simple rate of 0.25% and 0.2% per year under regular HVAC maintenance and suggest that these degradation rates are 1% and 0.5% without regular HVAC maintenance. They also emphasize that further work is needed to understand how to better quantify these degradations. It should be noted that in comparison with the building envelope,

1672

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

the efficiencies of mechanical system components are less dependent on age, as other factors, such as installation workmanship and maintenance have a more significant influence [35]. 3.2. Deterioration models Measurement of actual component performance degradation provides more information than simple failure time data. Faults or failures can be either defined as an ‘unacceptable performance loss’ or as an ‘unexpected malfunction’. Therefore modelling the performance degradation is more difficult since there are often several underlying processes [36]. Deterioration models can be classified into four categories models reflecting the deterioration conditions of the system: failure rate functions, Markov chains, Brownian motion models, and Gamma process models (Table 1). Deeper discussion of these deterioration models is available in the literature [37e39]. Failure rate functions are suitable for systems in which the degrading condition is assumed to be either one of two states: the functioning and failed state. The uncertainty in time of failure of a system or component is described by a service life distribution. Failure rate functions are especially useful for some mechanical and electrical engineering, like lamps. The main drawback of this method is that they cannot be measured for a particular component [39]. Markov deterioration models are useful when the conditions/ state of a system or component can be described by a limited number of categories. This method has been described briefly in Section 2.1. This method works well for visual inspections in assessing system conditions and is a widely applied model in civil infrastructures. However, it is not suitable when continuous measurement data represents gradual system deteriorations [38]. Furthermore the assumption of no memory (not dependent on the deterioration history) may be invalid for some cases [38]. When the system deterioration is measured on a continuous scale, two stochastic process models can be used to model this deterioration: Gamma process and Brownian motion [39,40]. A Gamma process model is a stochastic process model with a gamma distribution with non-negative increments, while the Brownian motion can be used to model increments and decrements. The Brownian motion should not be used when the variation in deteriorations is high [40]. Gamma process models are more suitable to represent gradual monotonic degradation over time due to wear, fatigue, corrosion, erosion etc. The gamma process with shape function (v(t) > 0) and scale parameter (u > 0) is a continuous-time stochastic process with independent gamma distributed increments.

Gaðxjv; uÞ ¼ uv xv1 expðuxÞ=GðvÞ

(1)

where G(a) is the gamma function for a > 0. According to van Noortwijk [39], the expected deterioration at time t can often be expressed with a power law

vðtÞ=u ¼ a t b

(2)

This process is called stationary at b ¼ 0. A gamma process with a non-linear shape function is called a non-stationary gamma when b s 1. The parameters of a gamma process (a, b, u in Eqs. (1) and (2)) can be estimated by using two common methods: maximum likelihood and method of moments [39]. An example of application of these two parameter estimation methods is available in literature [41]. This method has been successfully applied in time-based and condition-based maintenance. Thus far however Gamma process modelling has mainly been used for single components, not for complex systems [39]. 4. Other externalities This section discusses the impact of climate change on building energy use and building maintenance. 4.1. Climate change Predicted changes in the climate can be expected to impact building thermal performance during the building life cycle [42,43]. The results from climate change research usually are represented in terms of annual, seasonal or monthly changes. However, building performance simulation software typically uses weather files with hourly values to explore building behaviour under changing climate conditions. Four methods can be used for the conversion: dynamical downscaling, analogue scenarios, morphing; and weather generators [43,44]. Belcher et al. have suggested a morphing method to produce design weather data for building simulation based on data from the UK climate impacts programme (UKCIP02) [45]. Based on this morphing method, Jentsch et al. [46] have developed a weather file generator to create future hourly weather files for the UK and other countries, representing climate change. New UK climate change projections (UKCP09) were released in June 2009 and now provide probabilistic projections for a number of weather variables [12]. The UKCP09 dataset considers three types of uncertainty: natural internal climate variability, uncertainty in climate models, and uncertainty in future emission scenarios. It should be emphasized that the confidence levels in climate change projections are also dependent on the climatic variables themselves. For example, the confidence level in temperature is relatively higher than those in precipitation and cloudiness. UKCP09 also has a weather generator, which uses a statistical method to create projections of future daily or hourly climate values [47]. The main advantage of using this weather generator is that the inter-variable relationship between and within each of the variables is consistently maintained. Efforts have been reported that implement the UKCP09 weather generator into building thermal performance simulation. Tian et al. [48] compares

Table 1 Comparisons of four different methods for deterioration and maintenance models. Models

Advantages

Disadvantages

Descriptions

Failure rate function

Applied widely

Assume two states: functioning and failed.

Markov chain

Most commonly used approach, Suitable to incorporate information from visual inspection Suitable for systems subject to measurable deterioration

Failure rates cannot be measured for a particular component Not suitable for measurable deterioration on a continuous scale. Assumption of no memory Not suitable for monotonic deterioration

Brownian motion

Gamma process

Suitable for systems subject to measurable monotonic deterioration

More applied for single components rather for systems

Future chains dependent only on the present state and independent on any previous states (also called memory-less) A stochastic process with independent increments (or decrements) and a normal distribution A stochastic process with independent non-negative increments and a gamma distribution

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

three methods (FinkelsteineSchafer statistics, degree day theory, and metamodels) to reduce the computational demands. Jenkins et al. report work on simulation-generated regression to analyse the effects of different building designs under climate change projections [49]. Kershaw et al. suggest a method to consider the effects of an urban heat island under changing climate change as UKCP09 does not include this influence [50]. 4.2. Building maintenance Degradation of building components and engineering components will affect the overall building performance unless proper maintenance avoids degradation. As a principle, buildings start to degrade from the day they are put into service. In the British Standard BS 3811, building maintenance is defined as “the combination of all technical and associated administrative actions intended to retain an item in, or restore it to, a state in which it can perform its required function” [17]. Here the required function (or acceptable condition) will depend on the nature of the building, the importance of the element to the building use, and the maintenance policy of the user [51,52]. Building maintenance (Table 2) can be categorized into four methods: reactive, preventive, predictive, and reliability-centred maintenance [53]. Reactive maintenance is carried out when a fault (understood here as “unacceptable performance”) has occurred. This method seems low cost, but more money and time may be involved due to the unplanned downtime of equipment. Preventive maintenance is performed according to a time-based schedule and aims to restore the performance of equipment to an acceptable level before a fault occurs. Preventive maintenance may also help to extend the service life of equipment compared to equipment that only receives reactive maintenance. Predictive maintenance is based on intervention at a quantified material/equipment condition, prior to any significant degradation in the component state. This method can optimize the operation of the equipment, increase plant reliability, and minimize overtime cost. Predictive maintenance may provide a further saving of 8e12% over preventive maintenance [53]. However, further investment is needed to diagnose equipment conditions when this method is implemented. Finally, reliability-centred maintenance (RCM) is based on the operational requirements of specific plant components in relation to known reliability information. One important part of the process is to conduct a failure mode and effects analysis (FMEA) for the building components. NASA’s RCM guide provides reliability data for HVAC equipment in buildings [32]. Kwak et al. propose a method to predict an optimal maintenance model based on the reliability assessment and a Monte Carlo analysis for HVAC systems in office buildings in Japan [54]. 5. Methodology for longitudinal prediction of thermal building performance This section proposes a methodology to predict thermal performance of buildings over time-based on the methods

1673

reviewed in Section 2. The methodology is illustrated in Fig. 1, and consists of five steps. The first step is to collect data from historical sources such as publications, manufacturers, and maintenance records. The availability of this data determines which models can be used to estimate service life expectancy and performance deterioration of building components. In a building design phase, the life cycle thermal performance of buildings can be estimated by using relatively simple deterministic models. After the building is completed, more complex models may be developed that capitalise on actual data from building management systems and monitoring. The second step is to actually estimate the service life expectancy of the whole building and building elements. As described in Section 2.1, the service life can be predicted by two methods: deterministic and probabilistic methods. The British Building Research Establishment (BRE) [55] provides a method to estimate the service life and likelihood of replacement over the building life. For example, a building component might begin to be a candidate for replacement at year 10 if its reference service life is 20 years. At year 20, there is a chance of 50% that the component will have been replaced. Replacement will be completely finished (100% certainty) at year 30. This might be modelled using a Markov chain, but as explained in Section 2.1 this would require a lot of reliable data. Due to the limited reliable historic performance data available in many cases, a deterministic method may often be preferred. The third step is to predict the deterioration of thermal performance for the relevant building components. Two methods can be used: simple scenario and gamma process. As described in Section 3.2, a gamma process is suitable to model measurable monotonic deterioration. The efficiencies of boilers and chillers will degrade with time due to wear and tear or any other reasons. If sufficient field data is available, then the parameters required for a gamma process model can be obtained by using maximum likelihood and method of moments. In general however, due to the current lack of historic data on changes of building thermal performance over time, a simple scenario method as a first approximation is recommended. Such a simple scenario method allows estimating possible performance degradation for building components based on published generic data or work experience in facilities management. For example, simple annual degradation rates may be assumed to represent the deteriorations of efficiency in building components under different building maintenance scenarios [34]. The fourth step is to create a thermal simulation model of the building. There are many tools available that estimate the energy use of a building [56]. In most cases these tools evaluate buildings based on historical, average weather conditions from weather stations. However, long term prediction of building thermal performance can be taken to be more dependent on future climate conditions, not historical weather station data. Hence, annual weather files from climate change research may be also used to explore the life cycle thermal performance if possible. All of this seems to point to a preference for the use of advanced, transient buildings simulation programmes such as EnergyPlus, eQUEST, ESP-r, DOE-2, IES-VE and

Table 2 Comparisons of building maintenance methods. Methods

Advantages

Disadvantages

Descriptions

Reactive

Low cost, less staff

Run-to-failure maintenance

Preventive

Cost effective, may extend equipment life Reduction in labour cost and equipment downtime Most efficient maintenance, extend equipment life

Increased cost because of unexpected equipment downtime and overtime for maintenance Unplanned failure still possible, may have unnecessary maintenance More investment in equipment monitoring and staff training

Condition-based maintenance

High initial cost for equipment and training, need reliable historical data

Combination of preventive, predictive, and root cause failure analysis

Predictive Reliability-centred

Time-based maintenance

1674

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

6. Illustrative study 6.1. Case study using the deterministic method

Fig. 1. Schematic diagram of calculation of operating energy in a building life.

TRNSYS as described by the US DOE Energy Tool Directory [56]. Ideally these energy models will be calibrated based on actual measurements when the actual energy data becomes available. Reddy et al. provide a detailed review on the calibration of simulation models in building energy analysis and proposes a general methodology for calibrating simulation programs [57]. The sixth and final step is to predict building thermal performance as a function over service life by using building simulation software. Where a real building exists, and not just a simulation model, it is highly beneficial to use the data from a building management system for further calibration and updating of the thermal models. Note that analysis of longitudinal operational energy in buildings is not a stationary but a dynamic process. The energy model and input parameters thus can be refined over time. The following points should be borne in mind during the implementation of the procedure as described above: (1) The service life expectancy of building components does not always mean that the building component will be replaced after that period. It is also possible that a building may be abandoned due to obsolescence before the replacement moment is reached. The service life expectancy provides only a pointer to help designers and facility managers achieve optimum performance during the anticipated building life cycle. (2) This research focuses on thermal and energy performance of buildings and components. Note that the failure of these components may be due to other performance indicators, such as structural strength, noise level, aesthetics, and economy. These factors are considered in the second step of determining the service live expectancy although they are not explicitly included here. (3) Many other factors may also affect thermal performance of building components, such as improper installation, delayed maintenance, unexpected failures, and significant changes in occupant behaviour. Such factors are nearly impossible to take into account in estimating future thermal performance. These factors may be better addressed by implementing proper installation guidelines, building commissioning, and continuous monitoring of the status of crucial building components.

6.1.1. Building descriptions The approach to predict the performance of a building over time as described in the previous paragraph is demonstrated by means of a case study for a supermarket. This type of building is suitable as example because supermarket buildings have a high degree of repetition. They are mostly highly economical structures utilizing straightforward systems; most supermarket have their HVAC systems renewed after about 20 years. The changes in performance over the lifetime of such a building are then predicted taking into account a set of deterministic system degrading scenarios for the building systems, a climate change scenario, and HVAC system renewal after 20 years. Fig. 2 shows the building studied, which is one of a series of commercial benchmark models developed by US DOE (Department of Energy) [58]. This is a single-storey and six-zone building mainly for food sale and for the purpose of this article it is assumed to be located in Plymouth, UK. Packaged single zone air conditioner systems (PSZ-AC) in each thermal zone maintain the required temperatures within the indoor environment zones. Four compressor racks are employed to provide the refrigeration for eleven refrigeration cases in this supermarket and air-cooled condensers are used for the refrigeration. The HVAC equipment and systems are sized based on 99.6% heating (dry-bulb temperature 1.3  C) and 0.4% cooling (dry-bulb temperature 23.7  C). More detailed descriptions for this building are summarized in Table 3. The degradation of the building insulation and equipment efficiencies as assumed for the analysis are shown in Table 4. Here the thermal ageing of polyisocyanurate insulation is based on previous research [27], represented by means of the Agesim software and validated by measured data. The annual degradation rates in equipment efficiencies for fans, chillers, and gas furnaces are based on the data from literature [34]. It should be emphasized that the assumptions for deteriorations presented here are only for an indicative purpose. For detailed predictions more accurate degradation data may be obtained from manufacturers and through infield monitoring of real projects. More research, especially more reliable field measurement data, is urgently needed in this area. The operational energy modelling was carried out using EnergyPlus V5.0 [59]. One of the features in this program, Parametric Objects, has been used to automatically create multiple EnergyPlus input files in order to consider the change of the input parameters (Table 4) to predict annual energy use. In this case study, the overall building service life expectancy is assumed to be 40 years. As noted it is assumed that HVAC systems will be replaced once in year 20

Fig. 2. 3D view for a supermarket EnergyPlus model.

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680 Table 3 Input parameters of EnergyPlus model for a supermarket. Parameter

Descriptions

Location Building area Zone

Plymouth, UK 4180 m2 6 zones (sales, deli, office, dry storage, bakery, produce) (see Fig. 2) Mass wall, flat roof with insulation above deck, polyisocyanurate insulation (wall: 50 mm; roof: 127 mm, initial thermal conductivity: 0.021W/m K), 0.25 ACH infiltration, window only in south with window wall ratio 36%, window U value 3.24 W/m2 K and SHGC 0.385 Peak lights (office: 11.8 W/m2, dry storage: 8.6 W/m2, deli & sales & produce & bakery): 18.3 W/m2); peak equipment (office & dry storage: 8.07 W/m2, deli & bakery: 53.8 W/m2, sales & produce: 5.38 W/m2); peak gas plug (deli & bakery 26.9 W/m2); for detailed hourly schedules of these gains, please refer to [58]. PSZ-AC (packaged single zone air-conditioning), constant volume fan, gas heating coil, direct expansion (DX) cooling, initial gas burner efficiency (four coils: 0.78; two coils: 0.8), initial rated DX cooling COP (six coils: 3.74, 3.11, 3.19, 3.27, 3.11, 3.70), initial fan efficiency (six fans: 0.54, 0.57, 0.57, 0.59, 0.57, 0.57), heating set-point 21  C (6:00e22:00), cooling set-point 24  C (6:00e22:00) Two medium-temperature (rack A and C) and two low-temperature (rack B and D) compressor racks, eleven refrigeration cases, air-cooled condenser, initial COP (rack A & C: 2.5; rack B &D: 1.3)

Envelope

Internal gains

HVAC system

Refrigeration

because the service life expectancy of most of HVAC system is about these 20 years [17,20]. Future weather conditions (described by 100 climate files) for Plymouth, UK are used for both the time periods of the first 20 years (2020s, 2010e2039) and the second 20 years (2040s, 2030e2059). These climate files stem from UKCP09 predictions [10] based on the method described in [48]. For the second period of 20 years the efficiency of the HVAC equipment is assumed to have increase by 15% from initial efficiency (see Table 3). For this case study, only the medium emission scenario has been considered because the differences between three emission scenarios (high, medium, and low) are not significant for the 2020s and 2040s time periods. These emission scenarios represent the plausible future development of emissions of greenhouse gases determined by population change, socio-economic development, technology, etc. Fig. 3 shows the variations in annual mean temperature and heating degree days from 3000 weather files from the 2020s and 2040s, respectively. Simulation of a full year of climate data is performed every two years. Consequently, for the first and second 20 years’ timeframes, there are both 11 simulation models and every model needs to be run for

1675

a representative set of 100 weather files obtained from a total of 3000 weather files available in UKCP09 which are selected based on FinkelsteineSchafer statistics. So the total number of simulation runs needed is 4400. The University of Plymouth Condor Grid [60] has been used to expedite the calculations as thousands of simulation runs are involved in this type of analysis. 6.1.2. Results Figs. 4e6 show the predicted annual energy use for heating, cooling and refrigeration in this supermarket building over the full 40-year life cycle (first 20 years in the 2020s and second 20 years in the 2040s). According to the simulation results the number of unmet heating and cooling are both less than 60 h for all the models that have been investigated. As can be seen from Figs. 4e6, good maintenance can reduce the energy consumption compared to bad or no. However, annual energy use is still predicted to increase gradually due to the deterioration of building elements. For this case, the cooling and refrigeration energy would be reduced more significantly with time because of good maintenance in contrast with heating energy use. As can be expected, the annual heating energy is very likely to be reduced due to climate change from 2020s (first 20 years) to 2040s (second 20 years). Further technology improvements might add to this effect although there are many uncertain factors as shown in Fig. 4. Cooling energy is expected to remain stable between the first and second 20 years, see Fig. 5. Here the effects of the increased chiller COP are offset by the rising outdoor temperature. Fig. 6 indicates that the energy use for refrigeration is apparently being reduced. This is because the influence of climate change on refrigeration energy use is not significant compared to cooling energy consumption. Although this case study is only for illustrative purposes, it is obvious that annual operating energy use would significantly change due to building component degradation, building maintenance, and changing climate conditions. Obtaining a more accurate prediction requires availability of the reliable data from field measurements, monitoring, and manufacturers. 6.2. Case study using the stochastic method An illustrative example is given to show change of energy use by using probabilistic service life and degradation of a hypothetical component in a building. Firstly, cumulative distributions of service life and deterioration levels from gamma models are described. The results (change of energy use over the years) can be expressed through multi-scenario and probabilistic methods. The former approach is easier to understand, while the latter approach is more

Table 4 Degradation and replacement of building insulation and equipment performance. Parameters

Unit

Time

Values

Thermal conductivity

W/m K

2020sa

Gas burner efficiencyb

%

1/((0.000003  year5 þ 0.0001879  year4  0.0046597  year3 þ 0.0585943  year2  0.401224  year þ 6.8631878)  0.17611/0.0254) 0.0264 ini  (1  0.001  year) ini  (1  0.002  year) 1.15  ini  (1  0.001  year) 1.15  ini  (1  0.002  year) ini  (1  0.0025  year) ini  (1  0.01  year) 1.15  ini  (1  0.0025  year) 1.15  ini  (1  0.01  year) ini  (1  0.0025  year) ini  (1  0.01  year) 1.15  ini  (1  0.0025  year) 1.15  ini  (1  0.01  year) ini  (1  0.002  year) ini  (1  0.005  year) 1.15  ini  (1  0.002  year) 1.15  ini  (1  0.005  year)

Maintenance

DX cooling COPb

e b

Refrigeration Rack COP

e

Fan efficiencyb

%

a

2040s 2020s 2040s 2020s 2040s 2020s 2040s 2020s 2040s

Without or bad Maintenance

This formula is for the first 20 years (2020s). For the second 20 years (2040s), the thermal conductivity assumed to be constant. The efficiencies for gas heating coils, DX cooling, refrigeration, and fans for the second 20 years (2040s) are assumed to increase by 15% from initial efficiencies (see Table 3) because the old HVAC equipment will be replaced and the new system may have relative higher performance. b

1676

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

Fig. 3. Annual mean temperature (T) and annual heating degree days (HDD) with a base temperature of 15.5  C for the 2020s (2010e2039) and 2040s (2030e2059) time periods under medium emission scenarios in Plymouth from UKCP09.

appropriate when considering stochastic variations in the service life and degradation process of buildings. 6.2.1. Service life and degradation of building component Fig. 7 shows the service life and degradation for a hypothetical component in a building. As can be seen from the Fig. 7a, the

replacement of this component may start from the 9th year of use. The cumulative distribution function (CDF) can be obtained from the probabilistic method or engineering design method as described in Sections 2.1 and 2.2. Fig. 7b illustrates the deterioration level with the 5th and 95th percentile of this component. The parameters for this plot are taken

Fig. 4. Change of annual heating energy use in a supermarket for 40 years (the first 20 years in 2020s and the second 20 year in 2040s) with and without maintenance.

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

1677

Fig. 5. Change of annual cooling energy use in a supermarket for 40 years (the first 20 years in 2020s and the second 20 year in 2040s) with and without maintenance.

Fig. 6. Change of annual refrigerating energy use in a supermarket for 40 years (the first 20 years in 2020s and the second 20 year in 2040s) with and without maintenance.

1678

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

efficient equipment may be installed during major renovation. To explore the effects of variations in replacement time on energy use, the multi-scenario method can be used to show the change of energy use because of different replacement years, which should be based on Fig. 7a. Note that the energy use for different years can be expressed as probability distribution or cumulative distribution functions like the 5th and 15th year as shown in this figure. 6.2.3. Results of probabilistic methods Fig. 9 illustrate how to obtain the results by considering both probabilistic building service life and stochastic degradation for the building component. Firstly, the energy use over 20 years can be calculated when the replacement of this component occurred from the 10th to 20th year. Then the 11 different scenarios can be obtained, such as Fig. 9AeC. The probability of replacement at different years can be read from Fig. 7a, such as 3% probability at the 10th year. It is apparent that the energy use from the first to 9th year would be single cumulative distribution functions (see Fig. 9D and E) due to the temporal variations in gamma process model as it is not likely that the replacement would happen. After the 10th year, the energy use can be computed using the following equation:

SEnergy ¼

Fig. 7. Cumulative distribution function (CDF) of service life and degradation level with 5th and 95th percentile for a hypothetical component in a building.

as u 0.1176, a 0.2456, and b 2.8 for gamma model based on Eqs. (1) and (2) (see Section 3.2). Note that for every year, the curve of degradation can be expressed as cumulative distribution functions (CDFs) or probability distribution functions (PDFs) as shown in this figure for the 7th and 15th year. 6.2.2. Results of multi-scenario methods Fig. 8 shows the energy use due to degradation of this component, with replacement occurring in the 15th year. The energy use over the 15th year is less than the first year in this figure as more

X

Probability  CDFEnergy

(3)

Where, SEnergy is a weighted sum of energy use from different P scenarios, means the sum of different scenarios, Probability is the likelihood of replacement (see Fig. 7a), and CDFEnergy is the cumulative distribution function of energy at every scenario (i.e. different replacement year). For example, for the 10th year, there is a probability of 3% that the new component would replace the old one. Then the energy use at the 10th year would be a combination of 3% probability at use of a new component and 97% of probability still at the old one (Fig. 9F). Then for the 11th year, the energy use would be a combination of 3% probability happened at the 10th year, 6% probability happened at the 11th year, and 91% probability still at the old one. This process continues until the 20th year. Hence, the cumulative distribution functions (CDFs) of energy use at a specific year (Fig. 9H) would be similar to second order Monte Carlo analysis [61]. In Fig. 9H, the spread between these CDFs means the variations of energy use due to multi-scenario replacement years, while each CDF represents the temporal variations of degradation level at a given replacement year. A sampling-based Monte Carlo method, for example Latin hypercube sampling, can be used to get a single CDF based on Eq. (3) (like Fig. 9G and I). This method involves a lot of computation; metamodels may be used to speed up the calculation [48]. 7. Conclusion and remarks

Fig. 8. Change of energy use with 5th and 95th percentile due to degradation and replacement (assumed at the 15th year with upgrade) of a hypothetical component in a building.

This article proposes a methodology to predict possible changes in building performance during the building service life, which takes into account the change in physical properties due to deterioration of building systems and components, maintenance procedures and changing weather conditions. This methodology is applied to two simple case studies. Overall, this longitudinal prediction of building performance shows promise for building services and facility management. For instance it allows assessing when performance falls below a predefined threshold level and intervention is needed. However, at present the application of this method is very limited due to lack of reliable data. The approach presented in this paper is computationally intensive by nature, requiring a multitude of simulation runs (hundreds or thousands) to propagate uncertainties. However, advanced

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

1679

Fig. 9. Cumulative distribution functions of energy use with years as a result of probabilistic service life and degradation of a hypothetical component in a building.

computational environments like Grid Computing facilities already allow carrying out this type of analysis. It can be expected that trends towards further distributed computing (cloud computing) will make this an option for mainstream building analysts in the near future. Longitudinal prediction of building performance will offer these analysts a new deliverable towards their clients. A number of key areas have been identified where further research is needed to allow this method to go mainstream in building performance analysis studies: (1) This type of assessment needs to be based on extensive and reliable data, which may be obtained from published data, manufacturers, field measurements and monitoring. However, lack of relevant data remains one of the main constraints in implementing this type of longitudinal thermal performance analysis. Continuous monitoring of the whole building and components is necessary to provide more accurate historical data in order to create and update the models proposed in this paper. In addition, the manufacturers of building components should be encouraged to provide more data on the performance of their components over time, not just initial performance at start-up of the facility. (2) This article has explores both deterministic and stochastic views of deterioration parameters and replacement scenarios. The deterministic method is easy to understand and fast to compute. However, in real life, these aspects are subject to uncertainties. For example, the gamma process may be used to represent the temporal variability in the deterioration of building components. The variances of input parameters in Markov chain degradation models can be propagated using the Monte Carlo approach. The method proposed in this research may be further used to analyse economic impacts when considering both

maintenance policy and operational energy use in the long term by using deterministic and stochastic methods. (3) It is well known that most buildings are subject to a change of use somewhere during their life cycle. However, given the large impact of such a change of use on internal loads, comfort criteria and others, it seems not useful to express this in terms of uncertainties. Yet a number of scenarios might be worthwhile for investigation, like for instance a yearly overcrowding ratio, or trends in hotdesking where several users share building facilities. Thus far relatively few studies have focused on this topic. The basic protocols need to be developed to deal with such trends, and to study what the impact of likely scenarios might be. Acknowledgements The research described in this paper is funded by the Engineering and Physical Sciences Research Council (EPSRC) under grant EP/G000344/1 “Management of thermal performance risks in buildings subject to climate change”. We would thank the following researchers for their help of use Plymgrid: Dr Nicholas Outram, Dr. Alex Nimmo Smith and Dr George Grahsm, all from the University of Plymouth. References [1] Sartori I, Hestnes AG. Energy use in the life cycle of conventional and lowenergy buildings: a review article. Energy and Buildings 2007;39:249e57. [2] Ramesh T, Prakash R, Shukla KK. Life cycle energy analysis of buildings: an overview. Energy and Buildings 2010;42:1592e600. [3] Thormark C. A low energy building in a life cycle e its embodied energy, energy need for operation and recycling potential. Building and Environment 2002;37:429e35.

1680

P. de Wilde et al. / Building and Environment 46 (2011) 1670e1680

[4] Kneifel J. Life-cycle carbon and cost analysis of energy efficiency measures in new commercial buildings. Energy and Buildings 2010;42:333e40. [5] Brown N, Wright AJ, Shukla A, Stuart G. Longitudinal analysis of energy metering data from non-domestic buildings. Building Research and Information 2010;38:80e91. [6] Mathews EH, Botha CP. Improved thermal building management with the aid of integrated dynamic HVAC simulation. Building and Environment 2003;38:1423e9. [7] Chorier J, Hans J, Bazzana M, Chevalier J. Follow-up of in-use building components performances. Salford Quays, United Kingdom: CIB (Internatioal Council for Building) WORLD CONGRESS; 2010. Building a Better World. [8] Loy HM, John GA, Clements-Croome DJ, Fairey V, Neale K. Achieving quality through statistical prediction for building services systems. Building Services Engineering Research and Technology 2004;25:99e110. [9] Kirkham RJ, Alisa M, Pimenta da Silvva A, Grindley T, Brondsted J. EuroLifeForm: an integrated probabilistic whole life cycle cost and performance model for buildings and civil infrastructure. Proceedings of COBRA (Construction and Building Research Conference). Leeds: Leeds Metropolitan University; 2004. [10] Blom I, Itard L, Meijer A. LCA-based environmental assessment of the use and maintenance of heating and ventilation systems in Dutch dwellings. Building and Environment 2010;45:2362e72. [11] Wang X, Chen D, Ren Z. Assessment of climate change impact on residential building heating and cooling energy requirement in Australia. Building and Environment 2010;45:1663e82. [12] Murphy JM, Sexton DMH, Jenkins GJ, Booth BBB, Brown CC, Clark RT, et al. UK climate projections science report: climate change projections. Exeter (UK): Met Office Hadley Centre; 2009. [13] Boussabaine HA, Kirkham RL. Whole life-cycle costing: risk and risk responses. Oxford: Wiley-Blackwell; 2004. [14] Hovde PJ, Moser K. Performance based methods for service life prediction, state of the art reports. CIB Report (International Council for Building): Publication 294; 2004. [15] BCIS. Life expectancy of building components: surveyors’ experiences of buildings in use: a practical guide. 2nd rev. In: London: BCIS (Building cost information service), editor; 2006. [16] Langdon D. Life cycle costing (LCC) as a contribution to sustainable construction: a common methodology; 2007. Literature Review: European Commission. [17] CIBSE. Guide M maintenance engineering and management. London: Chartered Institute of Building Services Engineers; 2008. [18] BPG. BPG (Building Performance Group Ltd.) building fabric component life manual. London: Blackwell Science; 1999. [19] BPG. BPG (Building Performance Group) building services component life manual. Blackwell Science Ltd; 2001. [20] ASHRAE. HVAC applications handbook. Atlanta: American Society of Heating Refrigerating and Air-Conditioning; 2007. [21] Kirkham RJ, Boussabaine AH. Forecasting the residual service life of NHS hospital buildings: a stochastic approach. Construction Management and Economics 2005;23:521e9. [22] Augenbroe G, Park C. Towards a maintenance performance toolkit for GSA. Atlanta: Georgia Tech; 2002. GSA Funded Report. [23] Sharabah A, Setunge S, Zeephongsekul P. Use of Markov chain for deterioration modeling and risk management of infrastructure assets. ICIA: International Conference on Information and Automation; 2006. pp. 384e389. [24] Zhang Y, Augenbroe G, Vidakovic B. Uncertainty analysis in using Markov chain model to predict roof life cycle performance. Lyon (France): 10DBMC International Conference on durability of building materials and components; 2005. [25] Moser K, Edvardsen C. Engineering design methods for service life prediction. Rotterdam, Netherlands: 9th International Conference on Durability of Building Materials and Components DBMC; 2002. [26] BCO. Guide to specification: best practice in the specification for offices. London: BCO (British Council Office); 2009. [27] Singh SN, Coleman PD. Accelerated aging test methods for predicting the long term thermal resistance of closed-cell foam insulation. Center for Polyurethanes Industries Conference; 2007. [28] Graham M. Research reveals the LTTR method may be over-reporting results. Professional Roofing Magazine; 2006 January. [29] Mukhopadhyaya P, Bomberg MT, Kumaran MK, Drouin M, Lackey JC, Van Reenen D, et al. Long-term thermal resistance of polyisocyanurate foam insulation with gas barrier. Florida, USA: International Conference on Performance of Exterior Envelopes of Whole Buildings; December 5e10, 2004. [30] Pu-Europe. PU Europe technical report on thermal insulation materials made of rigid polyurethane foam (PUR/PIR). Federation of European Rigid Polyurethane Foam Associations; 2006. [31] Graham M. R-value concerns, R-values are found to be below LTTR. Professional Roofing Magazine; 2010 May. [32] NASA. Reliability-centered maintenance guide for facilities and collateral equipment. USA: National Aeronautics and Space Administration; 2008.

[33] Bannai M, Kimura Y, Fujii T, Sekiguchi K. Electric chiller performance evaluation based on actual operating data for energy service business. Okinawa, Japan: The International Conference on Electrical Engineering; 2008. [34] Griffith B, Long N, Torcellini P, Judkoff R, Crawley D, Ryan J. Methodology for modeling building energy performance across the commercial sector. NREL/ TP-550-41956. Golden, CO: National Renewable Energy Laboratory; 2008. [35] Piper JE. Handbook of facility assessment. Lilburn (USA): The Fairmont Press, Inc.; 2004. [36] Meeker WQ, Escobar LA. Statistical methods for reliability data. New York: Wiley; 1998. [37] Gorjian N, Ma L, Mittinty M, Yarlagadda P, Sun Y. A review on degradation models in reliability analysis; 2009. Proceedings of the 4th World Congress on Engineering Asset Management. Athens. [38] Frangopol DM, Kallen MJ, Van Noortwijk JM. Probabilistic models for life-cycle performance of deteriorating structures: review and future directions. Progress in Structural Engineering and Materials 2004;6:197e212. [39] van Noortwijk JM. A survey of the application of gamma processes in maintenance. Reliability Engineering & System Safety 2009;94:2e21. [40] Nicolai RP. Maintenance models for systems subject to measurable deterioration. Rozenberg Publishers; 2008. [41] Nicolai RP, Dekker R, van Noortwijk JM. A comparison of models for measurable deterioration: an application to coatings on steel structures. Reliability Engineering & System Safety 2007;92:1635e50. [42] de Wilde P, Tian W. The role of adaptive thermal comfort in the prediction of the thermal performance of a modern mixed-mode office building in the UK under climate change. Journal of Building Performance Simulation 2010;3:87e101. [43] Guan L. Preparation of future weather data to study the impact of climate change on buildings. Building and Environment 2009;44:793e800. [44] CIBSE. TM48 use of climate change scenarios for building simulation: the CIBSE future weather years. London: Chartered Institute of Building Services Engineers; 2009. [45] Belcher SE, Hacker JN, Powell DS. Constructing design weather data for future climates. Building Services Engineering Research and Technology 2005;26:49e61. [46] Jentsch MF, Bahaj AS, James PAB. Climate change future proofing of buildings e generation and assessment of building simulation weather files. Energy and Buildings 2008;40:2148e68. [47] Jones PD, Kilsby CG, Harpham C, Glenis V, Burton A. UK climate projections science report: projections of future daily climate for the UK from the weather generator. UK: University of Newcastle; 2009. [48] Tian W, de Wilde P. Thermal building simulation using the UKCP09 probabilistic climate projections. Journal of Building Performance Simulation, in press. doi:10.1080/19401493.2010.502246. [49] Jenkins D, Patidar S, Gibson G, Banfill P. Translating probabilistic climate predictions for use in building simulation. Proceedings of Conference: Adapting to Change: New Thinking on Comfort, Network for Comfort and Energy Use in Buildings. Windsor, UK: Cumberland Lodge; 2010. [50] Kershaw T, Sanderson M, Coley D, Eames M. Estimation of the urban heat island for UK climate change projections. Building Services Engineering Research and Technology 2010;31:251e63. [51] Chanter B, Swallow P. Building maintenance management. 2nd ed. Oxford: Blackwell; 2007. [52] Wordsworth P, Lee R. Lee’s building maintenance management. 4th ed. Malden, MA: Blackwell Science; 2001. [53] Sullivan GP, Pugh R, Melendez AP, Hunt WD. Operations & maintenance best practices guide 2.0 to achieving operational efficiency. Pacific Northwest National Laboratory for the Federal Energy Management Program US Department of Energy; 2004. [54] Kwak R-Y, Takakusagi A, Sohn J-Y, Fujii S, Park B-Y. Development of an optimal preventive maintenance model based on the reliability assessment for air-conditioning facilities in office buildings. Building and Environment 2004;39:1141e56. [55] Anderson J, Shiers D, Steele K. The green guide to specification. John Wiley and Sons; 2009. [56] US Department of Energy. Building energy software tools directory, http:// www.eere.energy.gov/buildings/tools_directory/, [accessed 01.09.10]. [57] Reddy AT, Maor I. ASHRAE RP-1051: procedures for reconciling computercalculated results with measured energy data. Atlanta, GA: American Society of Heating, Refrigerating, and Air-Conditioning Engineers; 2006. [58] Deru M, Field K, Studer D, Benne K, Griffith B, Torcellini P, et al. DOE commercial building research benchmarks for commercial buildings. Washington, DC: U.S. Department of Energy (DOE), Energy Efficiency and Renewable Energy, Office of Building Technologies; 2010. [59] US Department of Energy. EnergyPlus V5.0; 2010. [60] Condor Team. Condor high throughput computing; 2010. [61] de Wilde P, Tian W. Identification of key factors for uncertainty in the prediction of the thermal performance of an office building under climate change. Building Simulation 2009;2:157e74.